diff --git a/README.md b/README.md index b4358e0d..b28c2c8c 100644 --- a/README.md +++ b/README.md @@ -357,6 +357,7 @@ Qualcomm® AI Hub Models is licensed under BSD-3. See the [LICENSE file](../LICE | [DETR-ResNet50](https://aihub.qualcomm.com/models/detr_resnet50) | [qai_hub_models.models.detr_resnet50](qai_hub_models/models/detr_resnet50/README.md) | ✔️ | ✔️ | ✔️ | [DETR-ResNet50-DC5](https://aihub.qualcomm.com/models/detr_resnet50_dc5) | [qai_hub_models.models.detr_resnet50_dc5](qai_hub_models/models/detr_resnet50_dc5/README.md) | ✔️ | ✔️ | ✔️ | [MediaPipe-Face-Detection](https://aihub.qualcomm.com/models/mediapipe_face) | [qai_hub_models.models.mediapipe_face](qai_hub_models/models/mediapipe_face/README.md) | ✔️ | ✔️ | ✔️ +| [MediaPipe-Face-Detection-Quantized](https://aihub.qualcomm.com/models/mediapipe_face_quantized) | [qai_hub_models.models.mediapipe_face_quantized](qai_hub_models/models/mediapipe_face_quantized/README.md) | ✔️ | ✔️ | ✔️ | [MediaPipe-Hand-Detection](https://aihub.qualcomm.com/models/mediapipe_hand) | [qai_hub_models.models.mediapipe_hand](qai_hub_models/models/mediapipe_hand/README.md) | ✔️ | ✔️ | ✔️ | [YOLOv8-Detection](https://aihub.qualcomm.com/models/yolov8_det) | [qai_hub_models.models.yolov8_det](qai_hub_models/models/yolov8_det/README.md) | ✔️ | ✔️ | ✔️ | [YOLOv8-Detection-Quantized](https://aihub.qualcomm.com/models/yolov8_det_quantized) | [qai_hub_models.models.yolov8_det_quantized](qai_hub_models/models/yolov8_det_quantized/README.md) | ✔️ | ✔️ | ✔️ diff --git a/qai_hub_models/_version.py b/qai_hub_models/_version.py index cc0bf8d3..40f51364 100644 --- a/qai_hub_models/_version.py +++ b/qai_hub_models/_version.py @@ -2,4 +2,4 @@ # Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. # SPDX-License-Identifier: BSD-3-Clause # --------------------------------------------------------------------- -__version__ = "0.11.4" +__version__ = "0.12.0" diff --git a/qai_hub_models/models/_shared/llama/app.py b/qai_hub_models/models/_shared/llama/app.py index 0b08c9b3..0247b437 100644 --- a/qai_hub_models/models/_shared/llama/app.py +++ b/qai_hub_models/models/_shared/llama/app.py @@ -299,7 +299,9 @@ def generate_output_prompt( ) output_token = _get_tokens_from_logits(output) past_key_values = get_past_keyval_with_shift( - output[1:], num_of_past_key_heads=self.num_past_key_val_heads + output[1:], + past_key_start=0, + num_of_past_key_heads=self.num_past_key_val_heads, ).values() output_prompt = self.tokenizer.decode(output_token) print() diff --git a/qai_hub_models/models/_shared/llama/model.py b/qai_hub_models/models/_shared/llama/model.py index 4fc18b6f..09fd9d40 100644 --- a/qai_hub_models/models/_shared/llama/model.py +++ b/qai_hub_models/models/_shared/llama/model.py @@ -33,6 +33,7 @@ def get_past_key_names( start: int = 0, end: int = 8, num_of_past_key_heads=32, suffix="" ): past_key_val_name = [] + for i in range(start, end): cache_names = [ f"past_key_{i}_h{j}{suffix}" for j in range(num_of_past_key_heads) @@ -113,7 +114,10 @@ def load_input_cached_data( def get_past_keyval_with_shift( - past_key_vals: List[torch.Tensor], num_of_past_key_heads: int = 32 + past_key_vals: List[torch.Tensor], + past_key_start: int, + num_of_past_key_heads: int = 32, + new_key_suffix: str = "", ): """ Clip past key value to feed next iteration @@ -122,13 +126,18 @@ def get_past_keyval_with_shift( total_key_val = num_of_past_key_heads * 2 for i in range(0, len(past_key_vals), total_key_val): l_num = i // total_key_val + past_key_num = l_num + past_key_start for j, key in enumerate(past_key_vals[i : i + num_of_past_key_heads]): - tg_inputs[f"past_key_{l_num}_h{j}"] = key[:, :, :, 1:].detach() + tg_inputs[f"past_key_{past_key_num}_h{j}{new_key_suffix}"] = key[ + :, :, :, 1: + ].detach() for j, val in enumerate( past_key_vals[i + num_of_past_key_heads : i + total_key_val] ): - tg_inputs[f"past_value_{l_num}_h{j}"] = val[:, :, 1:, :].detach() + tg_inputs[f"past_value_{past_key_num}_h{j}{new_key_suffix}"] = val[ + :, :, 1:, : + ].detach() return tg_inputs @@ -244,12 +253,10 @@ def get_output_names( # Clipped hidden layers are named same as first part for all parts # Eventually, each split should have respective names. # layer_start, layer_end = get_hidden_layer_range_from_split(split_part=split_part, model_split_map=model_split_map) - layer_range = end - start - output_list = [ - output_name if output_name else f"layers_{layer_range - 1}_add_out_0" - ] + + output_list = [output_name if output_name else f"layers_{end - 1}_add_out_0"] output_list += get_past_key_names( - 0, layer_range, num_of_past_key_heads=past_key_val_heads, suffix="_out" + start, end, num_of_past_key_heads=past_key_val_heads, suffix="_out" ) return output_list diff --git a/qai_hub_models/models/aotgan/perf.yaml b/qai_hub_models/models/aotgan/perf.yaml index 913706fd..71d16bfa 100644 --- a/qai_hub_models/models/aotgan/perf.yaml +++ b/qai_hub_models/models/aotgan/perf.yaml @@ -45,11 +45,11 @@ models: - name: AOT-GAN performance_metrics: - torchscript_onnx_tflite: - inference_time: 153325.0 - throughput: 6.522093592043046 + inference_time: 152243.0 + throughput: 6.568446496719061 estimated_peak_memory_range: - min: 4362240 - max: 6399784 + min: 3362816 + max: 5464472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jwgod9915 + job_id: jw560rqn5 job_status: Passed torchscript_onnx_qnn: - inference_time: 153195.0 - throughput: 6.527628186298508 + inference_time: 152890.0 + throughput: 6.540650140623978 estimated_peak_memory_range: - min: 4284416 - max: 20967832 + min: 122880 + max: 25931488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 275 - job_id: jvgd6mm6p + job_id: jz5wr8m4p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,13 +81,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:59:30Z' + timestamp: '2024-08-27T01:11:08Z' - torchscript_onnx_tflite: - inference_time: 111811.0 - throughput: 8.943663861337436 + inference_time: 111545.0 + throughput: 8.964991707382671 estimated_peak_memory_range: - min: 2514944 - max: 194814400 + min: 3026944 + max: 194705552 primary_compute_unit: NPU precision: fp16 layer_info: @@ -95,14 +95,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: j1pv2nnzg + job_id: j1p3rxqmp job_status: Passed torchscript_onnx_qnn: - inference_time: 111783.0 - throughput: 8.945904117799666 + inference_time: 111698.0 + throughput: 8.95271177639707 estimated_peak_memory_range: min: 4235264 - max: 54437952 + max: 51365024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 275 - job_id: jz57o88ng + job_id: jmg9qk9mp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,13 +119,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:59:31Z' + timestamp: '2024-08-27T01:11:09Z' - torchscript_onnx_tflite: - inference_time: 152436.0 - throughput: 6.560130152982235 + inference_time: 152132.0 + throughput: 6.573239029264061 estimated_peak_memory_range: - min: 3223552 - max: 4823352 + min: 73728 + max: 471541752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -133,14 +133,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: j7gj3881p + job_id: jwgo9oe1g job_status: Passed torchscript_onnx_qnn: - inference_time: 97603.0 - throughput: 10.24558671352315 + inference_time: 92168.0 + throughput: 10.849752625640136 estimated_peak_memory_range: - min: 4325376 - max: 9056776 + min: 4464640 + max: 5769328 primary_compute_unit: NPU precision: fp16 layer_info: @@ -148,7 +148,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 275 - job_id: j0px0zz8p + job_id: jvgdm876g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -157,13 +157,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:59:33Z' + timestamp: '2024-08-27T01:11:11Z' - torchscript_onnx_tflite: - inference_time: 193513.0 - throughput: 5.167611478298616 + inference_time: 195261.0 + throughput: 5.121350397672859 estimated_peak_memory_range: - min: 32768 - max: 164807392 + min: 0 + max: 166374816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -171,14 +171,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jlpe6nn8g + job_id: j1pvnezzg job_status: Passed torchscript_onnx_qnn: - inference_time: 195249.0 - throughput: 5.121665155775446 + inference_time: 195436.0 + throughput: 5.116764567428723 estimated_peak_memory_range: - min: 4325376 - max: 45696896 + min: 3129344 + max: 43436800 primary_compute_unit: NPU precision: fp16 layer_info: @@ -186,7 +186,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 275 - job_id: jep2ozz6g + job_id: jo5mlov7g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -195,13 +195,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:59:38Z' + timestamp: '2024-08-27T01:11:15Z' - torchscript_onnx_tflite: - inference_time: 152989.0 - throughput: 6.536417650942225 + inference_time: 149443.0 + throughput: 6.6915144904746295 estimated_peak_memory_range: - min: 3162112 - max: 5062384 + min: 3268608 + max: 5593856 primary_compute_unit: NPU precision: fp16 layer_info: @@ -209,14 +209,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jygzz004g + job_id: j7gj8ok15 job_status: Passed torchscript_onnx_qnn: - inference_time: 97834.0 - throughput: 10.221395424903408 + inference_time: 92531.0 + throughput: 10.807188942084274 estimated_peak_memory_range: - min: 4456448 - max: 5648856 + min: 4472832 + max: 5724680 primary_compute_unit: NPU precision: fp16 layer_info: @@ -224,7 +224,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 275 - job_id: jo5m9ll7g + job_id: jz578kvnp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -233,13 +233,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:59:34Z' + timestamp: '2024-08-27T01:11:12Z' - torchscript_onnx_tflite: - inference_time: 152468.0 - throughput: 6.558753312170422 + inference_time: 152098.0 + throughput: 6.574708411681942 estimated_peak_memory_range: - min: 1613824 - max: 3500864 + min: 3289088 + max: 5883312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -247,14 +247,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jz5wyrr4g + job_id: jlpen848p job_status: Passed torchscript_onnx_qnn: - inference_time: 97427.0 - throughput: 10.264095168690405 + inference_time: 92888.0 + throughput: 10.765653259839807 estimated_peak_memory_range: - min: 4452352 - max: 5849920 + min: 4440064 + max: 5640384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -262,7 +262,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 275 - job_id: jegn1wwjp + job_id: jqp42mj2g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -271,13 +271,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:59:36Z' + timestamp: '2024-08-27T01:11:13Z' - torchscript_onnx_tflite: - inference_time: 152543.0 - throughput: 6.555528605049068 + inference_time: 149445.0 + throughput: 6.691424938940748 estimated_peak_memory_range: - min: 1654784 - max: 3282920 + min: 3489792 + max: 5474824 primary_compute_unit: NPU precision: fp16 layer_info: @@ -285,14 +285,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jmg9oqqmg + job_id: jygz08v45 job_status: Passed torchscript_onnx_qnn: - inference_time: 97399.0 - throughput: 10.26704586289387 + inference_time: 92465.0 + throughput: 10.814902936246147 estimated_peak_memory_range: - min: 4489216 - max: 6321656 + min: 4464640 + max: 5813384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -300,7 +300,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 275 - job_id: joprx77kp + job_id: j0pxz3e85 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -309,10 +309,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:59:37Z' + timestamp: '2024-08-27T01:11:14Z' - torchscript_onnx_qnn: - inference_time: 101222.0 - throughput: 9.879275256367192 + inference_time: 96156.0 + throughput: 10.399767045218187 estimated_peak_memory_range: min: 4202496 max: 4202496 @@ -323,7 +323,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 275 - job_id: jqp4e222g + job_id: jnp1m7qnp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -332,4 +332,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:59:32Z' + timestamp: '2024-08-27T01:11:10Z' diff --git a/qai_hub_models/models/convnext_tiny/perf.yaml b/qai_hub_models/models/convnext_tiny/perf.yaml index 243d8b35..2a5583af 100644 --- a/qai_hub_models/models/convnext_tiny/perf.yaml +++ b/qai_hub_models/models/convnext_tiny/perf.yaml @@ -45,11 +45,11 @@ models: - name: ConvNext-Tiny performance_metrics: - torchscript_onnx_tflite: - inference_time: 5646.0 - throughput: 177.11654268508678 + inference_time: 3257.0 + throughput: 307.0310101320233 estimated_peak_memory_range: - min: 28672 - max: 2387504 + min: 20480 + max: 2529288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 328 - job_id: j1pv2nmzg + job_id: jwgo9o11g job_status: Passed torchscript_onnx_qnn: - inference_time: 3806.0 - throughput: 262.7430373095113 + inference_time: 3793.0 + throughput: 263.6435539151068 estimated_peak_memory_range: - min: 45056 - max: 201053960 + min: 12288 + max: 190161728 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: jvgd6m46p + job_id: jnp1m7dnp job_status: Passed torchscript_onnx: - inference_time: 16310.0 - throughput: 61.31207847946045 + inference_time: 16336.0 + throughput: 61.21449559255632 estimated_peak_memory_range: - min: 45056 - max: 69049256 + min: 49152 + max: 69613664 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 189 - job_id: jqpy8y20g + job_id: jep2z476p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:58:40Z' + timestamp: '2024-08-27T01:10:21Z' - torchscript_onnx_tflite: - inference_time: 3975.0 - throughput: 251.57232704402514 + inference_time: 2406.0 + throughput: 415.6275976724855 estimated_peak_memory_range: min: 16384 - max: 218477744 + max: 212758544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 328 - job_id: j7gj38y1p + job_id: j1pvne1zg job_status: Passed torchscript_onnx_qnn: - inference_time: 2772.0 - throughput: 360.75036075036076 + inference_time: 2762.0 + throughput: 362.0564808110065 estimated_peak_memory_range: min: 618496 - max: 35335584 + max: 32475120 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: jz57o8nng + job_id: jvgdm8r6g job_status: Passed torchscript_onnx: - inference_time: 11590.0 - throughput: 86.28127696289906 + inference_time: 11528.0 + throughput: 86.74531575294934 estimated_peak_memory_range: - min: 663552 - max: 373270048 + min: 0 + max: 372090944 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 189 - job_id: j2p0oxx0p + job_id: jqpyyq40p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:58:41Z' + timestamp: '2024-08-27T01:10:22Z' - torchscript_onnx_tflite: - inference_time: 5594.0 - throughput: 178.7629603146228 + inference_time: 3226.0 + throughput: 309.98140111593307 estimated_peak_memory_range: - min: 65536 - max: 2320240 + min: 28672 + max: 1863832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 328 - job_id: jlpe6nx8g + job_id: j7gj8o015 job_status: Passed torchscript_onnx_qnn: - inference_time: 3543.0 - throughput: 282.2466836014677 + inference_time: 3407.0 + throughput: 293.51335485764605 estimated_peak_memory_range: - min: 647168 - max: 2212696 + min: 651264 + max: 2104736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: j0px0zr8p + job_id: jqp42mx2g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:58:34Z' + timestamp: '2024-08-27T01:10:16Z' - torchscript_onnx_tflite: - inference_time: 10883.0 - throughput: 91.88642837452909 + inference_time: 9028.0 + throughput: 110.76650420912716 estimated_peak_memory_range: - min: 45056 - max: 203222912 + min: 20480 + max: 204600672 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 328 - job_id: jygzz0y4g + job_id: jlpen8r8p job_status: Passed torchscript_onnx_qnn: - inference_time: 9615.0 - throughput: 104.00416016640666 + inference_time: 9639.0 + throughput: 103.74520178441747 estimated_peak_memory_range: - min: 675840 - max: 30674608 + min: 0 + max: 30654784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: jep2ozd6g + job_id: jopr7o4kg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:58:39Z' + timestamp: '2024-08-27T01:10:20Z' - torchscript_onnx_tflite: - inference_time: 5658.0 - throughput: 176.74089784376105 + inference_time: 3249.0 + throughput: 307.7870113881194 estimated_peak_memory_range: - min: 32768 - max: 2282480 + min: 40960 + max: 2359648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 328 - job_id: jz5wyrz4g + job_id: jygz08x45 job_status: Passed torchscript_onnx_qnn: - inference_time: 3538.0 - throughput: 282.6455624646693 + inference_time: 3417.0 + throughput: 292.654375182909 estimated_peak_memory_range: - min: 638976 - max: 2341392 + min: 655360 + max: 1962712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: jo5m9lk7g + job_id: j0pxz3785 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:58:35Z' + timestamp: '2024-08-27T01:10:17Z' - torchscript_onnx_tflite: - inference_time: 5651.0 - throughput: 176.95983011856308 + inference_time: 3227.0 + throughput: 309.88534242330337 estimated_peak_memory_range: - min: 28672 - max: 2663624 + min: 32768 + max: 2352752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 328 - job_id: jmg9oq2mg + job_id: jz5wr8d4p job_status: Passed torchscript_onnx_qnn: - inference_time: 3545.0 - throughput: 282.08744710860367 + inference_time: 3378.0 + throughput: 296.0331557134399 estimated_peak_memory_range: min: 634880 - max: 2136720 + max: 1925768 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: jegn1wqjp + job_id: jo5mlow7g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:58:36Z' + timestamp: '2024-08-27T01:10:18Z' - torchscript_onnx_tflite: - inference_time: 5625.0 - throughput: 177.77777777777777 + inference_time: 3237.0 + throughput: 308.9280197713933 estimated_peak_memory_range: - min: 57344 - max: 2406704 + min: 20480 + max: 3641440 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 328 - job_id: jnp1om1n5 + job_id: jmg9qk3mp job_status: Passed torchscript_onnx_qnn: - inference_time: 3673.0 - throughput: 272.2570106180234 + inference_time: 3521.0 + throughput: 284.0102243680773 estimated_peak_memory_range: - min: 638976 - max: 1898928 + min: 643072 + max: 2390000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: joprx7dkp + job_id: jegnwo9jg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:58:38Z' + timestamp: '2024-08-27T01:10:19Z' - torchscript_onnx_qnn: - inference_time: 3971.0 - throughput: 251.82573659027952 + inference_time: 3621.0 + throughput: 276.16680475006905 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: jqp4e242g + job_id: jz578kjnp job_status: Passed torchscript_onnx: - inference_time: 16970.0 - throughput: 58.927519151443725 + inference_time: 17017.0 + throughput: 58.76476464711759 estimated_peak_memory_range: - min: 61280256 - max: 61280256 + min: 62296064 + max: 62296064 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 189 - job_id: j1p8jkkq5 + job_id: j2p0xde0p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:58:42Z' + timestamp: '2024-08-27T01:10:23Z' diff --git a/qai_hub_models/models/convnext_tiny_w8a16_quantized/perf.yaml b/qai_hub_models/models/convnext_tiny_w8a16_quantized/perf.yaml index db9a59f5..19d1e118 100644 --- a/qai_hub_models/models/convnext_tiny_w8a16_quantized/perf.yaml +++ b/qai_hub_models/models/convnext_tiny_w8a16_quantized/perf.yaml @@ -45,11 +45,11 @@ models: - name: ConvNext-Tiny-w8a16-Quantized performance_metrics: - torchscript_onnx_qnn: - inference_time: 3488.0 - throughput: 286.697247706422 + inference_time: 3440.0 + throughput: 290.69767441860466 estimated_peak_memory_range: - min: 16384 - max: 127938288 + min: 12288 + max: 19895424 primary_compute_unit: NPU precision: int8 layer_info: @@ -57,7 +57,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jep2ozo6g + job_id: jz578kdnp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -66,13 +66,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:57:54Z' + timestamp: '2024-08-27T01:09:38Z' - torchscript_onnx_qnn: - inference_time: 2567.0 - throughput: 389.5597974289053 + inference_time: 2436.0 + throughput: 410.5090311986864 estimated_peak_memory_range: - min: 0 - max: 26877568 + min: 315392 + max: 34129392 primary_compute_unit: NPU precision: int8 layer_info: @@ -80,7 +80,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jqpy8y80g + job_id: jqp42mw2g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -89,13 +89,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:57:56Z' + timestamp: '2024-08-27T01:09:39Z' - torchscript_onnx_qnn: - inference_time: 3162.0 - throughput: 316.25553447185325 + inference_time: 3054.0 + throughput: 327.43942370661426 estimated_peak_memory_range: - min: 335872 - max: 1854232 + min: 339968 + max: 1781344 primary_compute_unit: NPU precision: int8 layer_info: @@ -103,7 +103,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: j1p8jkrq5 + job_id: jo5mloz7g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -112,13 +112,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:57:58Z' + timestamp: '2024-08-27T01:09:41Z' - torchscript_onnx_qnn: - inference_time: 4416.0 - throughput: 226.44927536231884 + inference_time: 4242.0 + throughput: 235.73785950023574 estimated_peak_memory_range: min: 315392 - max: 34503536 + max: 36728544 primary_compute_unit: NPU precision: int8 layer_info: @@ -126,7 +126,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jw56o0mn5 + job_id: jqpyyqd0p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -135,13 +135,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:58:03Z' + timestamp: '2024-08-27T01:09:45Z' - torchscript_onnx_qnn: - inference_time: 3159.0 - throughput: 316.5558721114277 + inference_time: 3054.0 + throughput: 327.43942370661426 estimated_peak_memory_range: - min: 327680 - max: 2026240 + min: 331776 + max: 1608744 primary_compute_unit: NPU precision: int8 layer_info: @@ -149,7 +149,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jogk6k0v5 + job_id: jegnwoejg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -158,13 +158,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:57:59Z' + timestamp: '2024-08-27T01:09:42Z' - torchscript_onnx_qnn: - inference_time: 3182.0 - throughput: 314.26775612822127 + inference_time: 3053.0 + throughput: 327.54667540124467 estimated_peak_memory_range: - min: 327680 - max: 1944656 + min: 335872 + max: 1919832 primary_compute_unit: NPU precision: int8 layer_info: @@ -172,7 +172,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jn5q4d1eg + job_id: jopr7oykg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -181,13 +181,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:58:00Z' + timestamp: '2024-08-27T01:09:43Z' - torchscript_onnx_qnn: - inference_time: 3178.0 - throughput: 314.6633102580239 + inference_time: 3060.0 + throughput: 326.797385620915 estimated_peak_memory_range: - min: 360448 - max: 1866192 + min: 335872 + max: 1581992 primary_compute_unit: NPU precision: int8 layer_info: @@ -195,7 +195,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: j1glwq82p + job_id: jep2z4m6p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -204,13 +204,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:58:02Z' + timestamp: '2024-08-27T01:09:44Z' - torchscript_onnx_qnn: - inference_time: 13268.0 - throughput: 75.3693096171239 + inference_time: 13515.0 + throughput: 73.99186089530151 estimated_peak_memory_range: - min: 389120 - max: 8788512 + min: 360448 + max: 8017712 primary_compute_unit: NPU precision: int8 layer_info: @@ -218,7 +218,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: j1p3or7mp + job_id: j2p0xd10p job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -227,13 +227,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:58:04Z' + timestamp: '2024-08-27T01:09:46Z' - torchscript_onnx_qnn: - inference_time: 3419.0 - throughput: 292.48318221702255 + inference_time: 3487.0 + throughput: 286.77946659019216 estimated_peak_memory_range: - min: 1523712 - max: 1523712 + min: 1503232 + max: 1503232 primary_compute_unit: NPU precision: int8 layer_info: @@ -241,7 +241,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: j2p0ox90p + job_id: j0pxz3185 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -250,4 +250,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:57:57Z' + timestamp: '2024-08-27T01:09:40Z' diff --git a/qai_hub_models/models/convnext_tiny_w8a8_quantized/perf.yaml b/qai_hub_models/models/convnext_tiny_w8a8_quantized/perf.yaml index f821dd78..01b4fb55 100644 --- a/qai_hub_models/models/convnext_tiny_w8a8_quantized/perf.yaml +++ b/qai_hub_models/models/convnext_tiny_w8a8_quantized/perf.yaml @@ -45,11 +45,11 @@ models: - name: ConvNext-Tiny-w8a8-Quantized performance_metrics: - torchscript_onnx_qnn: - inference_time: 1727.0 - throughput: 579.0387955993051 + inference_time: 1717.0 + throughput: 582.4111822947001 estimated_peak_memory_range: min: 16384 - max: 105409400 + max: 273666992 primary_compute_unit: NPU precision: int8 layer_info: @@ -57,7 +57,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: j1pv2n2zg + job_id: jogkkoyvg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -66,13 +66,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:57:01Z' + timestamp: '2024-08-27T01:08:50Z' - torchscript_onnx_qnn: - inference_time: 1197.0 - throughput: 835.421888053467 + inference_time: 1209.0 + throughput: 827.129859387924 estimated_peak_memory_range: - min: 163840 - max: 21384032 + min: 180224 + max: 21149152 primary_compute_unit: NPU precision: int8 layer_info: @@ -80,7 +80,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: j7gj3831p + job_id: jn5qdz2eg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -89,13 +89,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:57:04Z' + timestamp: '2024-08-27T01:08:51Z' - torchscript_onnx_qnn: - inference_time: 1661.0 - throughput: 602.0469596628537 + inference_time: 1653.0 + throughput: 604.9606775559589 estimated_peak_memory_range: - min: 180224 - max: 1426776 + min: 184320 + max: 1452960 primary_compute_unit: NPU precision: int8 layer_info: @@ -103,7 +103,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jygzz0z4g + job_id: jw560r1n5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -112,13 +112,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:57:06Z' + timestamp: '2024-08-27T01:08:53Z' - torchscript_onnx_qnn: inference_time: 2155.0 throughput: 464.0371229698376 estimated_peak_memory_range: - min: 0 - max: 24032576 + min: 163840 + max: 25814608 primary_compute_unit: NPU precision: int8 layer_info: @@ -126,7 +126,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jvgd6m66p + job_id: j7gj8ol15 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -135,13 +135,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:57:11Z' + timestamp: '2024-08-27T01:08:57Z' - torchscript_onnx_qnn: - inference_time: 1675.0 - throughput: 597.0149253731344 + inference_time: 1666.0 + throughput: 600.2400960384153 estimated_peak_memory_range: - min: 180224 - max: 1735040 + min: 176128 + max: 1287672 primary_compute_unit: NPU precision: int8 layer_info: @@ -149,7 +149,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jz5wyry4g + job_id: j1p3rxmmp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -158,13 +158,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:57:07Z' + timestamp: '2024-08-27T01:08:54Z' - torchscript_onnx_qnn: - inference_time: 1672.0 - throughput: 598.0861244019138 + inference_time: 1670.0 + throughput: 598.8023952095808 estimated_peak_memory_range: - min: 184320 - max: 1503352 + min: 180224 + max: 1825640 primary_compute_unit: NPU precision: int8 layer_info: @@ -172,7 +172,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jmg9oqomg + job_id: jwgo9ov1g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -181,13 +181,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:57:09Z' + timestamp: '2024-08-27T01:08:55Z' - torchscript_onnx_qnn: - inference_time: 1668.0 - throughput: 599.5203836930456 + inference_time: 1670.0 + throughput: 598.8023952095808 estimated_peak_memory_range: - min: 180224 - max: 1786416 + min: 176128 + max: 1840072 primary_compute_unit: NPU precision: int8 layer_info: @@ -195,7 +195,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jnp1omon5 + job_id: j1pvnewzg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -204,13 +204,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:57:10Z' + timestamp: '2024-08-27T01:08:56Z' - torchscript_onnx_qnn: - inference_time: 6589.0 - throughput: 151.76809834572774 + inference_time: 6587.0 + throughput: 151.8141794443601 estimated_peak_memory_range: - min: 208896 - max: 8897296 + min: 225280 + max: 8129056 primary_compute_unit: NPU precision: int8 layer_info: @@ -218,7 +218,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jz57o8ong + job_id: jlpen8v8p job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -227,13 +227,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:57:13Z' + timestamp: '2024-08-27T01:08:58Z' - torchscript_onnx_qnn: - inference_time: 1801.0 - throughput: 555.247084952804 + inference_time: 1856.0 + throughput: 538.7931034482758 estimated_peak_memory_range: - min: 462848 - max: 462848 + min: 450560 + max: 450560 primary_compute_unit: NPU precision: int8 layer_info: @@ -241,7 +241,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 215 - job_id: jlpe6n68g + job_id: j1glqok2p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -250,4 +250,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:57:05Z' + timestamp: '2024-08-27T01:08:52Z' diff --git a/qai_hub_models/models/ddrnet23_slim/perf.yaml b/qai_hub_models/models/ddrnet23_slim/perf.yaml index 6c344e63..81dfba81 100644 --- a/qai_hub_models/models/ddrnet23_slim/perf.yaml +++ b/qai_hub_models/models/ddrnet23_slim/perf.yaml @@ -45,11 +45,11 @@ models: - name: DDRNet23-Slim performance_metrics: - torchscript_onnx_tflite: - inference_time: 5138.0 - throughput: 194.62826002335538 + inference_time: 5146.0 + throughput: 194.32568985619898 estimated_peak_memory_range: - min: 1024000 - max: 3446080 + min: 45056 + max: 2335576 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 131 - job_id: j1p3orznp + job_id: j2p0xd2np job_status: Passed torchscript_onnx: - inference_time: 9618.0 - throughput: 103.97171969224371 + inference_time: 9585.0 + throughput: 104.32968179447053 estimated_peak_memory_range: - min: 14008320 - max: 225285464 + min: 9863168 + max: 14098408 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 155 - job_id: jqp4e2y2g + job_id: jvgdm8ozg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,13 +81,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:56:11Z' + timestamp: '2024-08-27T01:08:03Z' - torchscript_onnx_tflite: - inference_time: 3536.0 - throughput: 282.80542986425337 + inference_time: 3528.0 + throughput: 283.4467120181406 estimated_peak_memory_range: - min: 12288 - max: 73321808 + min: 987136 + max: 75652176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -95,14 +95,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 131 - job_id: jwgod9lk5 + job_id: j1p8k6mop job_status: Passed torchscript_onnx: - inference_time: 6096.0 - throughput: 164.04199475065616 + inference_time: 6023.0 + throughput: 166.03021749958492 estimated_peak_memory_range: - min: 11841536 - max: 93788528 + min: 11804672 + max: 91319648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 155 - job_id: j0px0zl8p + job_id: jz5wr8q4p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,13 +119,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:56:12Z' + timestamp: '2024-08-27T01:08:04Z' - torchscript_onnx_tflite: - inference_time: 5155.0 - throughput: 193.98642095053347 + inference_time: 5063.0 + throughput: 197.51135690302192 estimated_peak_memory_range: - min: 1028096 - max: 2584240 + min: 995328 + max: 5505312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -133,7 +133,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 131 - job_id: j1pv2nlrg + job_id: jogkkoqng job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,13 +142,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:55:55Z' + timestamp: '2024-08-27T01:07:51Z' - torchscript_onnx_tflite: - inference_time: 7583.0 - throughput: 131.87392852433075 + inference_time: 7505.0 + throughput: 133.24450366422386 estimated_peak_memory_range: - min: 1015808 - max: 62945472 + min: 999424 + max: 62094096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -156,7 +156,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 131 - job_id: j7gj38rep + job_id: jn5qdzrog job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,13 +165,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:55:56Z' + timestamp: '2024-08-27T01:07:51Z' - torchscript_onnx_tflite: - inference_time: 5164.0 - throughput: 193.64833462432222 + inference_time: 5252.0 + throughput: 190.4036557501904 estimated_peak_memory_range: - min: 2142208 - max: 4652944 + min: 1024000 + max: 2744992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -179,7 +179,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 131 - job_id: jlpe6n7vg + job_id: j1glqo3mp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,13 +188,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:55:57Z' + timestamp: '2024-08-27T01:07:52Z' - torchscript_onnx_tflite: - inference_time: 5124.0 - throughput: 195.160031225605 + inference_time: 5206.0 + throughput: 192.0860545524395 estimated_peak_memory_range: - min: 1015808 - max: 105668656 + min: 1019904 + max: 131228552 primary_compute_unit: NPU precision: fp16 layer_info: @@ -202,7 +202,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 131 - job_id: jygzz0lxg + job_id: jw560rny5 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,13 +211,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:55:58Z' + timestamp: '2024-08-27T01:07:53Z' - torchscript_onnx_tflite: - inference_time: 5190.0 - throughput: 192.67822736030828 + inference_time: 5083.0 + throughput: 196.7342120794806 estimated_peak_memory_range: - min: 2138112 - max: 4662784 + min: 1036288 + max: 6763864 primary_compute_unit: NPU precision: fp16 layer_info: @@ -225,7 +225,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 131 - job_id: jz5wyrlmg + job_id: j1p3rxenp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -234,10 +234,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:56:00Z' + timestamp: '2024-08-27T01:07:54Z' - torchscript_onnx: - inference_time: 9587.0 - throughput: 104.30791697089809 + inference_time: 9722.0 + throughput: 102.85949393128986 estimated_peak_memory_range: min: 9859072 max: 9859072 @@ -248,7 +248,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 155 - job_id: jo5m9l07g + job_id: jmg9qkwmp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -257,4 +257,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:56:13Z' + timestamp: '2024-08-27T01:08:05Z' diff --git a/qai_hub_models/models/deeplabv3_plus_mobilenet/perf.yaml b/qai_hub_models/models/deeplabv3_plus_mobilenet/perf.yaml index 0beccb5c..91f81e89 100644 --- a/qai_hub_models/models/deeplabv3_plus_mobilenet/perf.yaml +++ b/qai_hub_models/models/deeplabv3_plus_mobilenet/perf.yaml @@ -45,11 +45,11 @@ models: - name: DeepLabV3-Plus-MobileNet performance_metrics: - torchscript_onnx_tflite: - inference_time: 13533.0 - throughput: 73.89344565137073 + inference_time: 13214.0 + throughput: 75.67731194187982 estimated_peak_memory_range: - min: 21024768 - max: 23321608 + min: 22151168 + max: 24319880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 98 - job_id: j1pv2ndrg + job_id: jogkko7ng job_status: Passed torchscript_onnx_qnn: - inference_time: 12902.0 - throughput: 77.50736319950396 + inference_time: 12964.0 + throughput: 77.13668620796051 estimated_peak_memory_range: - min: 4210688 - max: 17886808 + min: 4468736 + max: 16787432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 124 - job_id: jvgd6m3zp + job_id: j7gj8o2e5 job_status: Passed torchscript_onnx: - inference_time: 17863.0 - throughput: 55.981638022728546 + inference_time: 17899.0 + throughput: 55.86904296329404 estimated_peak_memory_range: - min: 44273664 - max: 56959952 + min: 48726016 + max: 61432184 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jqpy8yxlg + job_id: jqp42md1g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:55:29Z' + timestamp: '2024-08-27T01:07:26Z' - torchscript_onnx_tflite: - inference_time: 9518.0 - throughput: 105.06408909434755 + inference_time: 9526.0 + throughput: 104.97585555322276 estimated_peak_memory_range: - min: 21827584 - max: 96427024 + min: 21913600 + max: 96911168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 98 - job_id: j7gj387ep + job_id: jn5qdzeog job_status: Passed torchscript_onnx_qnn: - inference_time: 9407.0 - throughput: 106.30381630700542 + inference_time: 9399.0 + throughput: 106.39429726566657 estimated_peak_memory_range: - min: 3174400 - max: 29099248 + min: 3186688 + max: 29571920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 124 - job_id: jz57o849g + job_id: jlpen8wvp job_status: Passed torchscript_onnx: - inference_time: 13647.0 - throughput: 73.27617791455998 + inference_time: 13884.0 + throughput: 72.02535292422932 estimated_peak_memory_range: - min: 49897472 - max: 127731984 + min: 48930816 + max: 125711376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j2p0oxjnp + job_id: j0pxz36l5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:55:30Z' + timestamp: '2024-08-27T01:07:27Z' - torchscript_onnx_tflite: - inference_time: 13110.0 - throughput: 76.27765064836004 + inference_time: 12975.0 + throughput: 77.07129094412332 estimated_peak_memory_range: - min: 22188032 - max: 23828904 + min: 18845696 + max: 22492160 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 98 - job_id: jlpe6nzvg + job_id: j1glqo2mp job_status: Passed torchscript_onnx_qnn: - inference_time: 12552.0 - throughput: 79.66857871255577 + inference_time: 12032.0 + throughput: 83.11170212765957 estimated_peak_memory_range: - min: 3227648 - max: 29694016 + min: 3235840 + max: 4782432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 124 - job_id: j0px0z4lp + job_id: jz5wr83mp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:55:23Z' + timestamp: '2024-08-27T01:07:21Z' - torchscript_onnx_tflite: - inference_time: 18313.0 - throughput: 54.606017583137664 + inference_time: 18135.0 + throughput: 55.141990625861595 estimated_peak_memory_range: - min: 22142976 - max: 99904880 + min: 22151168 + max: 97556576 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 98 - job_id: jygzz0mxg + job_id: jw560rzy5 job_status: Passed torchscript_onnx_qnn: - inference_time: 18909.0 - throughput: 52.88486963879634 + inference_time: 18404.0 + throughput: 54.33601391001956 estimated_peak_memory_range: - min: 3194880 - max: 30448800 + min: 3239936 + max: 32794752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 124 - job_id: jep2ozwqg + job_id: jz578kl9p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:55:28Z' + timestamp: '2024-08-27T01:07:25Z' - torchscript_onnx_tflite: - inference_time: 13223.0 - throughput: 75.62580352416245 + inference_time: 13093.0 + throughput: 76.37668983426258 estimated_peak_memory_range: - min: 22163456 - max: 24496128 + min: 22142976 + max: 28930608 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 98 - job_id: jz5wyr7mg + job_id: j1p3rx1np job_status: Passed torchscript_onnx_qnn: - inference_time: 12507.0 - throughput: 79.95522507395859 + inference_time: 12148.0 + throughput: 82.31807704972012 estimated_peak_memory_range: - min: 3244032 - max: 4534920 + min: 3239936 + max: 4584424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 124 - job_id: jo5m9lm9g + job_id: jmg9qky8p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:55:24Z' + timestamp: '2024-08-27T01:07:22Z' - torchscript_onnx_tflite: - inference_time: 13165.0 - throughput: 75.9589821496392 + inference_time: 13039.0 + throughput: 76.69299792928905 estimated_peak_memory_range: - min: 22151168 - max: 240741832 + min: 22138880 + max: 24493024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 98 - job_id: jmg9oqm8g + job_id: jwgo9onkg job_status: Passed torchscript_onnx_qnn: - inference_time: 12552.0 - throughput: 79.66857871255577 + inference_time: 12119.0 + throughput: 82.51505899826718 estimated_peak_memory_range: - min: 3231744 - max: 29943496 + min: 3203072 + max: 4569544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 124 - job_id: jegn1wnqp + job_id: jnp1m7w7p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:55:26Z' + timestamp: '2024-08-27T01:07:23Z' - torchscript_onnx_tflite: - inference_time: 13302.0 - throughput: 75.17666516313336 + inference_time: 13043.0 + throughput: 76.66947788085564 estimated_peak_memory_range: - min: 22126592 - max: 274626480 + min: 22151168 + max: 43495344 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 98 - job_id: jnp1omj75 + job_id: j1pvnerrg job_status: Passed torchscript_onnx_qnn: - inference_time: 12400.0 - throughput: 80.64516129032258 + inference_time: 12144.0 + throughput: 82.34519104084322 estimated_peak_memory_range: - min: 3219456 - max: 4686752 + min: 3239936 + max: 4605968 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 124 - job_id: joprx707p + job_id: jvgdm8qzg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:55:27Z' + timestamp: '2024-08-27T01:07:24Z' - torchscript_onnx_qnn: - inference_time: 12973.0 - throughput: 77.08317274339012 + inference_time: 12497.0 + throughput: 80.01920460910618 estimated_peak_memory_range: min: 3170304 max: 3170304 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 124 - job_id: jqp4e211g + job_id: jygz08jx5 job_status: Passed torchscript_onnx: - inference_time: 16767.0 - throughput: 59.64096141229797 + inference_time: 16579.0 + throughput: 60.31726883406719 estimated_peak_memory_range: - min: 69464064 - max: 69464064 + min: 69480448 + max: 69480448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1p8jkxo5 + job_id: jo5mlo69g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:55:31Z' + timestamp: '2024-08-27T01:07:28Z' diff --git a/qai_hub_models/models/deeplabv3_plus_mobilenet_quantized/perf.yaml b/qai_hub_models/models/deeplabv3_plus_mobilenet_quantized/perf.yaml index 9c3742a7..a82b3178 100644 --- a/qai_hub_models/models/deeplabv3_plus_mobilenet_quantized/perf.yaml +++ b/qai_hub_models/models/deeplabv3_plus_mobilenet_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: DeepLabV3-Plus-MobileNet-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 3321.0 - throughput: 301.11412225233363 + inference_time: 3313.0 + throughput: 301.84123151222457 estimated_peak_memory_range: min: 12288 - max: 17418128 + max: 146159848 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: jwgod98k5 + job_id: j1p8k62op job_status: Passed torchscript_onnx_qnn: - inference_time: 5185.0 - throughput: 192.86403085824494 + inference_time: 5161.0 + throughput: 193.7608990505716 estimated_peak_memory_range: - min: 20480 - max: 14628712 + min: 811008 + max: 12314968 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jz57o899g + job_id: jlpen8evp job_status: Passed torchscript_onnx: - inference_time: 16177.0 - throughput: 61.816158743895656 + inference_time: 16102.0 + throughput: 62.10408644888834 estimated_peak_memory_range: - min: 51585024 - max: 55069392 + min: 43241472 + max: 50272032 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 51 total_layers: 171 - job_id: j1p8jk1o5 + job_id: jo5mloq9g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:54:45Z' + timestamp: '2024-08-27T01:06:47Z' - torchscript_onnx_tflite: - inference_time: 2475.0 - throughput: 404.04040404040404 + inference_time: 2464.0 + throughput: 405.84415584415586 estimated_peak_memory_range: min: 12288 - max: 61945568 + max: 63772880 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: j1pv2n7rg + job_id: jogkkovng job_status: Passed torchscript_onnx_qnn: - inference_time: 3827.0 - throughput: 261.30128037627384 + inference_time: 3847.0 + throughput: 259.94281258123215 estimated_peak_memory_range: min: 802816 - max: 30666496 + max: 32629040 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jqp4e231g + job_id: jygz08ox5 job_status: Passed torchscript_onnx: - inference_time: 12225.0 - throughput: 81.79959100204499 + inference_time: 12035.0 + throughput: 83.09098462816785 estimated_peak_memory_range: - min: 48271360 - max: 115930048 + min: 49840128 + max: 119787984 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 51 total_layers: 171 - job_id: jogk6k8n5 + job_id: jegnwolqg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:54:46Z' + timestamp: '2024-08-27T01:06:48Z' - torchscript_onnx_tflite: - inference_time: 3317.0 - throughput: 301.4772384684956 + inference_time: 3279.0 + throughput: 304.9710277523635 estimated_peak_memory_range: min: 12288 - max: 131897288 + max: 1718688 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: j7gj38qep + job_id: jn5qdz0og job_status: Passed torchscript_onnx_qnn: - inference_time: 3998.0 - throughput: 250.12506253126563 + inference_time: 3933.0 + throughput: 254.25883549453343 estimated_peak_memory_range: - min: 847872 - max: 2416632 + min: 823296 + max: 2076784 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jo5m9l89g + job_id: jmg9qkj8p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:54:38Z' + timestamp: '2024-08-27T01:06:41Z' - torchscript_onnx_tflite: - inference_time: 4111.0 - throughput: 243.24981756263682 + inference_time: 4142.0 + throughput: 241.42926122646065 estimated_peak_memory_range: - min: 12288 - max: 63694192 + min: 217088 + max: 64088896 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: jlpe6nyvg + job_id: j1glqo6mp job_status: Passed torchscript_onnx_qnn: - inference_time: 5599.0 - throughput: 178.6033220217896 + inference_time: 5620.0 + throughput: 177.93594306049823 estimated_peak_memory_range: - min: 802816 - max: 31400208 + min: 819200 + max: 32781456 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jqpy8ymlg + job_id: jqp42mk1g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:54:43Z' + timestamp: '2024-08-27T01:06:45Z' - torchscript_onnx_tflite: - inference_time: 3312.0 - throughput: 301.9323671497585 + inference_time: 3305.0 + throughput: 302.571860816944 estimated_peak_memory_range: min: 12288 - max: 1669432 + max: 5710208 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: jygzz0nxg + job_id: jw560rey5 job_status: Passed torchscript_onnx_qnn: - inference_time: 4031.0 - throughput: 248.07740014884644 + inference_time: 3943.0 + throughput: 253.613999492772 estimated_peak_memory_range: - min: 851968 - max: 2371160 + min: 835584 + max: 2102608 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jegn1wkqp + job_id: jnp1m7y7p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:54:39Z' + timestamp: '2024-08-27T01:06:42Z' - torchscript_onnx_tflite: - inference_time: 3314.0 - throughput: 301.75015087507546 + inference_time: 3318.0 + throughput: 301.38637733574444 estimated_peak_memory_range: - min: 12288 - max: 144967560 + min: 5591040 + max: 7705856 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: jz5wyr4mg + job_id: j1p3rxvnp job_status: Passed torchscript_onnx_qnn: - inference_time: 4031.0 - throughput: 248.07740014884644 + inference_time: 3968.0 + throughput: 252.01612903225808 estimated_peak_memory_range: - min: 851968 - max: 2097608 + min: 823296 + max: 2358456 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: joprx7w7p + job_id: jvgdm8ezg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:54:40Z' + timestamp: '2024-08-27T01:06:43Z' - torchscript_onnx_tflite: - inference_time: 3320.0 - throughput: 301.2048192771084 + inference_time: 3295.0 + throughput: 303.49013657056145 estimated_peak_memory_range: - min: 32768 - max: 3437008 + min: 12288 + max: 152254896 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: jmg9oqd8g + job_id: jwgo9okkg job_status: Passed torchscript_onnx_qnn: - inference_time: 4049.0 - throughput: 246.9745616201531 + inference_time: 3958.0 + throughput: 252.65285497726126 estimated_peak_memory_range: - min: 831488 - max: 2354016 + min: 815104 + max: 2387712 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jep2ozeqg + job_id: jz578k09p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:54:42Z' + timestamp: '2024-08-27T01:06:44Z' - torchscript_onnx_tflite: - inference_time: 14911.0 - throughput: 67.06458319361545 + inference_time: 14735.0 + throughput: 67.8656260604004 estimated_peak_memory_range: - min: 5537792 - max: 46398256 + min: 5562368 + max: 48657888 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: jnp1om675 + job_id: j1pvne0rg job_status: Passed torchscript_onnx_qnn: - inference_time: 18377.0 - throughput: 54.41584589432443 + inference_time: 18821.0 + throughput: 53.13213963126295 estimated_peak_memory_range: - min: 835584 - max: 8540704 + min: 905216 + max: 9037808 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: j2p0ox6np + job_id: j0pxz3nl5 job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:54:44Z' + timestamp: '2024-08-27T01:06:46Z' - torchscript_onnx_tflite: - inference_time: 128551.0 - throughput: 7.779013776633398 + inference_time: 127584.0 + throughput: 7.837973413594181 estimated_peak_memory_range: - min: 11485184 - max: 65634176 + min: 10756096 + max: 28887176 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 3 layers_on_cpu: 0 total_layers: 104 - job_id: jvgd6m2zp + job_id: j7gj8oze5 job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,10 +406,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:54:33Z' + timestamp: '2024-08-27T01:06:37Z' - torchscript_onnx_qnn: - inference_time: 4356.0 - throughput: 229.5684113865932 + inference_time: 4253.0 + throughput: 235.12814483893723 estimated_peak_memory_range: min: 802816 max: 802816 @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: j0px0zxlp + job_id: jz5wr82mp job_status: Passed torchscript_onnx: - inference_time: 17392.0 - throughput: 57.49770009199632 + inference_time: 14736.0 + throughput: 67.86102062975027 estimated_peak_memory_range: - min: 74272768 - max: 74272768 + min: 74305536 + max: 74305536 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 51 total_layers: 171 - job_id: jn5q4dvog + job_id: jopr7o87g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:54:47Z' + timestamp: '2024-08-27T01:06:49Z' diff --git a/qai_hub_models/models/deeplabv3_resnet50/perf.yaml b/qai_hub_models/models/deeplabv3_resnet50/perf.yaml index e5dd1304..2a4e3a84 100644 --- a/qai_hub_models/models/deeplabv3_resnet50/perf.yaml +++ b/qai_hub_models/models/deeplabv3_resnet50/perf.yaml @@ -45,11 +45,11 @@ models: - name: DeepLabV3-ResNet50 performance_metrics: - torchscript_onnx_tflite: - inference_time: 292954.0 - throughput: 3.413505191941397 + inference_time: 290843.0 + throughput: 3.4382811344952433 estimated_peak_memory_range: - min: 15785984 - max: 196981112 + min: 12288 + max: 149300160 primary_compute_unit: GPU precision: fp16 layer_info: @@ -57,7 +57,7 @@ models: layers_on_gpu: 95 layers_on_cpu: 0 total_layers: 95 - job_id: jz57o879g + job_id: jlpen8ovp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -66,13 +66,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:53:36Z' + timestamp: '2024-08-27T01:05:46Z' - torchscript_onnx_tflite: - inference_time: 227499.0 - throughput: 4.395623717027328 + inference_time: 205420.0 + throughput: 4.868075163080518 estimated_peak_memory_range: - min: 21639168 - max: 48971312 + min: 21024768 + max: 48086416 primary_compute_unit: GPU precision: fp16 layer_info: @@ -80,7 +80,7 @@ models: layers_on_gpu: 95 layers_on_cpu: 0 total_layers: 95 - job_id: jqp4e291g + job_id: jygz082x5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -89,13 +89,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:53:37Z' + timestamp: '2024-08-27T01:05:47Z' - torchscript_onnx_tflite: - inference_time: 290413.0 - throughput: 3.4433720253569917 + inference_time: 294293.0 + throughput: 3.3979741278249906 estimated_peak_memory_range: - min: 0 - max: 148239064 + min: 4096 + max: 177961048 primary_compute_unit: GPU precision: fp16 layer_info: @@ -103,7 +103,7 @@ models: layers_on_gpu: 95 layers_on_cpu: 0 total_layers: 95 - job_id: j0px0zdlp + job_id: jz5wr8wmp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -112,13 +112,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:53:38Z' + timestamp: '2024-08-27T01:05:48Z' - torchscript_onnx_tflite: - inference_time: 748558.0 - throughput: 1.3359018272465193 + inference_time: 486137.0 + throughput: 2.057033305426248 estimated_peak_memory_range: - min: 22282240 - max: 52742496 + min: 22265856 + max: 54838512 primary_compute_unit: GPU precision: fp16 layer_info: @@ -126,7 +126,7 @@ models: layers_on_gpu: 95 layers_on_cpu: 0 total_layers: 95 - job_id: jo5m9ld9g + job_id: jmg9qk08p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -135,13 +135,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:53:39Z' + timestamp: '2024-08-27T01:05:48Z' - torchscript_onnx_tflite: - inference_time: 290292.0 - throughput: 3.444807297479779 + inference_time: 290878.0 + throughput: 3.437867422080735 estimated_peak_memory_range: - min: 0 - max: 148457536 + min: 16384 + max: 147760272 primary_compute_unit: GPU precision: fp16 layer_info: @@ -149,7 +149,7 @@ models: layers_on_gpu: 95 layers_on_cpu: 0 total_layers: 95 - job_id: jegn1w7qp + job_id: jnp1m727p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -158,13 +158,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:53:41Z' + timestamp: '2024-08-27T01:05:49Z' - torchscript_onnx_tflite: - inference_time: 290341.0 - throughput: 3.4442259274439366 + inference_time: 290240.0 + throughput: 3.4454244762954795 estimated_peak_memory_range: min: 0 - max: 147822872 + max: 307211088 primary_compute_unit: GPU precision: fp16 layer_info: @@ -172,7 +172,7 @@ models: layers_on_gpu: 95 layers_on_cpu: 0 total_layers: 95 - job_id: joprx7n7p + job_id: jvgdm8nzg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -181,13 +181,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:53:42Z' + timestamp: '2024-08-27T01:05:50Z' - torchscript_onnx_tflite: - inference_time: 291682.0 - throughput: 3.428391193148703 + inference_time: 291047.0 + throughput: 3.4358711823176327 estimated_peak_memory_range: min: 0 - max: 321577016 + max: 279665056 primary_compute_unit: GPU precision: fp16 layer_info: @@ -195,7 +195,7 @@ models: layers_on_gpu: 95 layers_on_cpu: 0 total_layers: 95 - job_id: jep2ozvqg + job_id: jz578k29p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -204,4 +204,4 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:53:43Z' + timestamp: '2024-08-27T01:05:51Z' diff --git a/qai_hub_models/models/densenet121/perf.yaml b/qai_hub_models/models/densenet121/perf.yaml index 9d75e7dd..7bc45ec3 100644 --- a/qai_hub_models/models/densenet121/perf.yaml +++ b/qai_hub_models/models/densenet121/perf.yaml @@ -45,11 +45,11 @@ models: - name: DenseNet-121 performance_metrics: - torchscript_onnx_tflite: - inference_time: 1930.0 - throughput: 518.1347150259068 + inference_time: 1911.0 + throughput: 523.2862375719518 estimated_peak_memory_range: - min: 20480 - max: 205426288 + min: 28672 + max: 1702512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 312 - job_id: j0px0z3lp + job_id: jnp1m748p job_status: Passed torchscript_onnx_qnn: - inference_time: 1992.0 - throughput: 502.00803212851406 + inference_time: 1990.0 + throughput: 502.51256281407035 estimated_peak_memory_range: - min: 16384 - max: 39641592 + min: 286720 + max: 40582560 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 372 - job_id: j1p8jk6o5 + job_id: jqp42ml1g job_status: Passed torchscript_onnx: - inference_time: 1914.0 - throughput: 522.466039707419 + inference_time: 1919.0 + throughput: 521.1047420531527 estimated_peak_memory_range: min: 12288 - max: 18053856 + max: 18126288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 374 - job_id: j7gj38wep + job_id: jn5qdz6og job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:53:06Z' + timestamp: '2024-08-27T01:05:17Z' - torchscript_onnx_tflite: - inference_time: 1279.0 - throughput: 781.8608287724785 + inference_time: 1280.0 + throughput: 781.25 estimated_peak_memory_range: min: 16384 - max: 102418080 + max: 102659584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 312 - job_id: jo5m9lo9g + job_id: jvgdm8xrg job_status: Passed torchscript_onnx_qnn: - inference_time: 1319.0 - throughput: 758.1501137225171 + inference_time: 1325.0 + throughput: 754.7169811320755 estimated_peak_memory_range: min: 0 - max: 21065920 + max: 21255440 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 372 - job_id: jogk6kon5 + job_id: j0pxz3kl5 job_status: Passed torchscript_onnx: - inference_time: 1338.0 - throughput: 747.3841554559043 + inference_time: 1321.0 + throughput: 757.002271006813 estimated_peak_memory_range: min: 0 - max: 106089376 + max: 106856528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 374 - job_id: jlpe6nlvg + job_id: j1glqo4mp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:53:07Z' + timestamp: '2024-08-27T01:05:18Z' - torchscript_onnx_tflite: - inference_time: 1937.0 - throughput: 516.2622612287042 + inference_time: 1915.0 + throughput: 522.1932114882507 estimated_peak_memory_range: - min: 32768 - max: 2019288 + min: 36864 + max: 2344712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 312 - job_id: jegn1woqp + job_id: jz5wr8emp job_status: Passed torchscript_onnx_qnn: - inference_time: 1782.0 - throughput: 561.1672278338945 + inference_time: 1781.0 + throughput: 561.4823133071309 estimated_peak_memory_range: min: 630784 - max: 1938456 + max: 1897904 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 372 - job_id: j1glwq1mp + job_id: jegnwo6qg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:53:00Z' + timestamp: '2024-08-27T01:05:12Z' - torchscript_onnx_tflite: - inference_time: 2638.0 - throughput: 379.07505686125853 + inference_time: 2608.0 + throughput: 383.4355828220859 estimated_peak_memory_range: - min: 12288 - max: 102063248 + min: 32768 + max: 102970224 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 312 - job_id: joprx7o7p + job_id: jmg9qkl8p job_status: Passed torchscript_onnx_qnn: - inference_time: 2716.0 - throughput: 368.1885125184094 + inference_time: 2679.0 + throughput: 373.2736095558044 estimated_peak_memory_range: - min: 794624 - max: 20121760 + min: 618496 + max: 23720656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 372 - job_id: j1pv2n9rg + job_id: jogkkoeng job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:53:05Z' + timestamp: '2024-08-27T01:05:16Z' - torchscript_onnx_tflite: - inference_time: 1951.0 - throughput: 512.557662737058 + inference_time: 1913.0 + throughput: 522.7391531625718 estimated_peak_memory_range: - min: 12288 - max: 14575384 + min: 24576 + max: 4922928 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 312 - job_id: jep2oz4qg + job_id: jnp1m747p job_status: Passed torchscript_onnx_qnn: - inference_time: 1796.0 - throughput: 556.7928730512249 + inference_time: 1788.0 + throughput: 559.2841163310962 estimated_peak_memory_range: - min: 634880 - max: 2216016 + min: 638976 + max: 2319176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 372 - job_id: jw56o0dy5 + job_id: jopr7ov7g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:53:01Z' + timestamp: '2024-08-27T01:05:13Z' - torchscript_onnx_tflite: - inference_time: 1944.0 - throughput: 514.40329218107 + inference_time: 1909.0 + throughput: 523.8344683080147 estimated_peak_memory_range: - min: 20480 - max: 7276056 + min: 45056 + max: 1929856 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 312 - job_id: jqpy8yqlg + job_id: jvgdm8xzg job_status: Passed torchscript_onnx_qnn: - inference_time: 1789.0 - throughput: 558.9714924538848 + inference_time: 1796.0 + throughput: 556.7928730512249 estimated_peak_memory_range: - min: 630784 - max: 1852480 + min: 638976 + max: 2028704 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 372 - job_id: j1p3orwnp + job_id: jqpyyq1lp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:53:02Z' + timestamp: '2024-08-27T01:05:14Z' - torchscript_onnx_tflite: - inference_time: 1937.0 - throughput: 516.2622612287042 + inference_time: 1926.0 + throughput: 519.2107995846313 estimated_peak_memory_range: - min: 16384 - max: 15947032 + min: 40960 + max: 9158216 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 312 - job_id: j2p0oxdnp + job_id: jz578ky9p job_status: Passed torchscript_onnx_qnn: - inference_time: 1938.0 - throughput: 515.9958720330237 + inference_time: 1912.0 + throughput: 523.0125523012553 estimated_peak_memory_range: min: 634880 - max: 2289088 + max: 1977888 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 372 - job_id: jwgod94k5 + job_id: j2p0xdznp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:53:04Z' + timestamp: '2024-08-27T01:05:15Z' - torchscript_onnx_qnn: - inference_time: 1992.0 - throughput: 502.00803212851406 + inference_time: 2024.0 + throughput: 494.0711462450593 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 372 - job_id: jn5q4dzog + job_id: jo5mlon9g job_status: Passed torchscript_onnx: - inference_time: 2007.0 - throughput: 498.2561036372696 + inference_time: 2043.0 + throughput: 489.47626040137055 estimated_peak_memory_range: - min: 18337792 - max: 18337792 + min: 19333120 + max: 19333120 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 374 - job_id: jygzz04xg + job_id: jw560r2y5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:53:08Z' + timestamp: '2024-08-27T01:05:19Z' diff --git a/qai_hub_models/models/detr_resnet101/perf.yaml b/qai_hub_models/models/detr_resnet101/perf.yaml index 12dbd211..7bbf5f32 100644 --- a/qai_hub_models/models/detr_resnet101/perf.yaml +++ b/qai_hub_models/models/detr_resnet101/perf.yaml @@ -45,26 +45,26 @@ models: - name: DETR-ResNet101 performance_metrics: - torchscript_onnx_tflite: - inference_time: 24837.0 - throughput: 40.262511575472075 + inference_time: 17302.0 + throughput: 57.79678649867068 estimated_peak_memory_range: - min: 2568192 - max: 10729824 + min: 73728 + max: 2802560 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 855 - layers_on_gpu: 1 + layers_on_npu: 856 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: j1p8jk8k5 + job_id: jz5wr863p job_status: Passed torchscript_onnx: - inference_time: 22270.0 - throughput: 44.9034575662326 + inference_time: 22185.0 + throughput: 45.075501464953796 estimated_peak_memory_range: - min: 57344 - max: 133831672 + min: 2813952 + max: 5963896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jz5wyr8mg + job_id: jwgo9o6qg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,28 +81,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:52:25Z' + timestamp: '2024-08-27T01:04:41Z' - torchscript_onnx_tflite: - inference_time: 17532.0 - throughput: 57.03855806525211 + inference_time: 12341.0 + throughput: 81.03071063933231 estimated_peak_memory_range: - min: 811008 - max: 290141584 + min: 73728 + max: 307617296 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 855 - layers_on_gpu: 1 + layers_on_npu: 856 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jogk6kdw5 + job_id: jmg9qknwp job_status: Passed torchscript_onnx: - inference_time: 15643.0 - throughput: 63.926356836923866 + inference_time: 15605.0 + throughput: 64.08202499198974 estimated_peak_memory_range: - min: 2154496 - max: 256163008 + min: 1282048 + max: 253292912 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jmg9oqk8g + job_id: j7gj8ovv5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,21 +119,21 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:52:26Z' + timestamp: '2024-08-27T01:04:42Z' - torchscript_onnx_tflite: - inference_time: 24815.0 - throughput: 40.298206729800526 + inference_time: 17166.0 + throughput: 58.25468950250495 estimated_peak_memory_range: - min: 2576384 - max: 9440560 + min: 90112 + max: 3716280 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 855 - layers_on_gpu: 1 + layers_on_npu: 856 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jn5q4dwng + job_id: jnp1m7z8p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,13 +142,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:52:10Z' + timestamp: '2024-08-27T01:04:28Z' - torchscript_onnx_tflite: - inference_time: 33612.0 - throughput: 29.751279305010115 + inference_time: 23775.0 + throughput: 42.06098843322818 estimated_peak_memory_range: - min: 2633728 - max: 232813616 + min: 0 + max: 240771648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -156,7 +156,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: j1glwq7jp + job_id: jvgdm81rg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,21 +165,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:52:11Z' + timestamp: '2024-08-27T01:04:29Z' - torchscript_onnx_tflite: - inference_time: 24809.0 - throughput: 40.30795275907936 + inference_time: 17291.0 + throughput: 57.83355502862761 estimated_peak_memory_range: - min: 16384 - max: 7508648 + min: 86016 + max: 2197296 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 855 - layers_on_gpu: 1 + layers_on_npu: 856 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jw56o0v65 + job_id: jz578krvp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,21 +188,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:52:12Z' + timestamp: '2024-08-27T01:04:30Z' - torchscript_onnx_tflite: - inference_time: 24806.0 - throughput: 40.31282754172378 + inference_time: 17311.0 + throughput: 57.76673791231009 estimated_peak_memory_range: - min: 2588672 - max: 9184648 + min: 86016 + max: 2565008 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 855 - layers_on_gpu: 1 + layers_on_npu: 856 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: j1p3or83p + job_id: jqp42mr8g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,21 +211,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:52:13Z' + timestamp: '2024-08-27T01:04:31Z' - torchscript_onnx_tflite: - inference_time: 24723.0 - throughput: 40.44816567568661 + inference_time: 17366.0 + throughput: 57.58378440631118 estimated_peak_memory_range: - min: 2551808 - max: 10628888 + min: 77824 + max: 2539576 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 855 - layers_on_gpu: 1 + layers_on_npu: 856 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jwgod9mq5 + job_id: j0pxz3o35 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -234,13 +234,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:52:14Z' + timestamp: '2024-08-27T01:04:32Z' - torchscript_onnx: - inference_time: 22813.0 - throughput: 43.83465567877964 + inference_time: 23088.0 + throughput: 43.31254331254331 estimated_peak_memory_range: - min: 123736064 - max: 123736064 + min: 123715584 + max: 123715584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -248,7 +248,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jnp1om775 + job_id: jlpen8dop job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -257,4 +257,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:52:28Z' + timestamp: '2024-08-27T01:04:42Z' diff --git a/qai_hub_models/models/detr_resnet101_dc5/perf.yaml b/qai_hub_models/models/detr_resnet101_dc5/perf.yaml index 3e56cd56..37793820 100644 --- a/qai_hub_models/models/detr_resnet101_dc5/perf.yaml +++ b/qai_hub_models/models/detr_resnet101_dc5/perf.yaml @@ -45,26 +45,26 @@ models: - name: DETR-ResNet101-DC5 performance_metrics: - torchscript_onnx_tflite: - inference_time: 149910.0 - throughput: 6.670669068107531 + inference_time: 122288.0 + throughput: 8.177417244537486 estimated_peak_memory_range: - min: 9527296 - max: 17071656 + min: 212992 + max: 3369080 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 856 - layers_on_gpu: 1 + layers_on_npu: 857 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 857 - job_id: j1glwqejp + job_id: jvgdm0lrg job_status: Passed torchscript_onnx: - inference_time: 124629.0 - throughput: 8.023814681976106 + inference_time: 125102.0 + throughput: 7.993477322504836 estimated_peak_memory_range: - min: 155648 - max: 134575480 + min: 40960 + max: 134382048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jo5m9l4dg + job_id: j1p3r823p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,28 +81,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:51:34Z' + timestamp: '2024-08-27T01:03:53Z' - torchscript_onnx_tflite: - inference_time: 106429.0 - throughput: 9.395935318381268 + inference_time: 92457.0 + throughput: 10.815838714213093 estimated_peak_memory_range: - min: 7897088 - max: 503707968 + min: 176128 + max: 503243744 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 856 - layers_on_gpu: 1 + layers_on_npu: 857 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 857 - job_id: jw56o0q65 + job_id: jz57863vp job_status: Passed torchscript_onnx: - inference_time: 94514.0 - throughput: 10.580443108957404 + inference_time: 95282.0 + throughput: 10.495161730442266 estimated_peak_memory_range: - min: 0 - max: 458244688 + min: 1986560 + max: 462844416 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: jegn1wxkp + job_id: jwgo9mqqg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,21 +119,21 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:51:35Z' + timestamp: '2024-08-27T01:03:53Z' - torchscript_onnx_tflite: - inference_time: 140262.0 - throughput: 7.129514765225079 + inference_time: 123539.0 + throughput: 8.094609799334624 estimated_peak_memory_range: - min: 9535488 - max: 16636992 + min: 24576 + max: 141292896 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 856 - layers_on_gpu: 1 + layers_on_npu: 857 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 857 - job_id: j1p3orq3p + job_id: jqp42808g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,13 +142,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:51:19Z' + timestamp: '2024-08-27T01:03:40Z' - torchscript_onnx_tflite: - inference_time: 165496.0 - throughput: 6.042442113404554 + inference_time: 139949.0 + throughput: 7.145460131905194 estimated_peak_memory_range: - min: 9629696 - max: 446251728 + min: 36864 + max: 434178096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -156,7 +156,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 857 - job_id: jwgod9eq5 + job_id: j0pxzm235 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,21 +165,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:51:20Z' + timestamp: '2024-08-27T01:03:41Z' - torchscript_onnx_tflite: - inference_time: 149943.0 - throughput: 6.669200963032619 + inference_time: 120987.0 + throughput: 8.265350822815675 estimated_peak_memory_range: - min: 9572352 - max: 16143472 + min: 167936 + max: 2388672 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 856 - layers_on_gpu: 1 + layers_on_npu: 857 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 857 - job_id: j1pv2n4kg + job_id: jo5ml4ydg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,21 +188,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:51:21Z' + timestamp: '2024-08-27T01:03:42Z' - torchscript_onnx_tflite: - inference_time: 147306.0 - throughput: 6.788589738367752 + inference_time: 125434.0 + throughput: 7.9723201045968395 estimated_peak_memory_range: - min: 9560064 - max: 17885464 + min: 188416 + max: 2303192 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 856 - layers_on_gpu: 1 + layers_on_npu: 857 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 857 - job_id: j7gj381vp + job_id: jegnwx8kg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,21 +211,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:51:22Z' + timestamp: '2024-08-27T01:03:43Z' - torchscript_onnx_tflite: - inference_time: 139625.0 - throughput: 7.162041181736795 + inference_time: 128685.0 + throughput: 7.770913470878502 estimated_peak_memory_range: - min: 9580544 - max: 16551336 + min: 16384 + max: 3470200 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 856 - layers_on_gpu: 1 + layers_on_npu: 857 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 857 - job_id: jlpe6n2og + job_id: jopr79j0g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -234,13 +234,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:51:24Z' + timestamp: '2024-08-27T01:03:44Z' - torchscript_onnx: - inference_time: 124620.0 - throughput: 8.024394158241053 + inference_time: 124774.0 + throughput: 8.014490198278487 estimated_peak_memory_range: - min: 127164416 - max: 127164416 + min: 127340544 + max: 127340544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -248,7 +248,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 856 - job_id: joprx790p + job_id: j1pvnekkg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -257,4 +257,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:51:37Z' + timestamp: '2024-08-27T01:03:54Z' diff --git a/qai_hub_models/models/detr_resnet50/perf.yaml b/qai_hub_models/models/detr_resnet50/perf.yaml index 1e0be3b6..43e48a0e 100644 --- a/qai_hub_models/models/detr_resnet50/perf.yaml +++ b/qai_hub_models/models/detr_resnet50/perf.yaml @@ -45,26 +45,26 @@ models: - name: DETR-ResNet50 performance_metrics: - torchscript_onnx_tflite: - inference_time: 20889.0 - throughput: 47.87208578677773 + inference_time: 14073.0 + throughput: 71.0580544304697 estimated_peak_memory_range: - min: 2711552 - max: 9745616 + min: 106496 + max: 3448680 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 787 - layers_on_gpu: 1 + layers_on_npu: 788 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 788 - job_id: jwgod91q5 + job_id: jvgdm09rg job_status: Passed torchscript_onnx: - inference_time: 16353.0 - throughput: 61.15085916957133 + inference_time: 16348.0 + throughput: 61.169562025935896 estimated_peak_memory_range: min: 40960 - max: 99980800 + max: 100971744 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 737 - job_id: jep2oz3rg + job_id: j1p3r833p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,28 +81,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:50:43Z' + timestamp: '2024-08-27T01:03:05Z' - torchscript_onnx_tflite: - inference_time: 14495.0 - throughput: 68.98930665746809 + inference_time: 9936.0 + throughput: 100.64412238325282 estimated_peak_memory_range: - min: 2105344 - max: 237664704 + min: 90112 + max: 252255952 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 787 - layers_on_gpu: 1 + layers_on_npu: 788 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 788 - job_id: j1pv2nzkg + job_id: jz5786wvp job_status: Passed torchscript_onnx: - inference_time: 11339.0 - throughput: 88.19119851838786 + inference_time: 11370.0 + throughput: 87.95074758135443 estimated_peak_memory_range: - min: 2936832 - max: 203215920 + min: 0 + max: 201758976 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 737 - job_id: jqpy8yv8g + job_id: jwgo9m0qg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,21 +119,21 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:50:44Z' + timestamp: '2024-08-27T01:03:05Z' - torchscript_onnx_tflite: - inference_time: 20753.0 - throughput: 48.18580446200549 + inference_time: 14085.0 + throughput: 70.99751508697196 estimated_peak_memory_range: - min: 2564096 - max: 9240432 + min: 61440 + max: 2484792 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 787 - layers_on_gpu: 1 + layers_on_npu: 788 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 788 - job_id: j7gj38kvp + job_id: jqp428o8g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,13 +142,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:50:28Z' + timestamp: '2024-08-27T01:02:52Z' - torchscript_onnx_tflite: - inference_time: 27408.0 - throughput: 36.48569760653824 + inference_time: 16816.0 + throughput: 59.467174119885826 estimated_peak_memory_range: - min: 2719744 - max: 201129280 + min: 90112 + max: 209079280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -156,7 +156,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 788 - job_id: jlpe6n4og + job_id: j0pxzmj35 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,21 +165,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:50:29Z' + timestamp: '2024-08-27T01:02:53Z' - torchscript_onnx_tflite: - inference_time: 20958.0 - throughput: 47.714476572192005 + inference_time: 14105.0 + throughput: 70.89684509039348 estimated_peak_memory_range: - min: 2555904 - max: 9986240 + min: 77824 + max: 2512312 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 787 - layers_on_gpu: 1 + layers_on_npu: 788 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 788 - job_id: jygzz0vog + job_id: jo5ml42dg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,21 +188,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:50:30Z' + timestamp: '2024-08-27T01:02:54Z' - torchscript_onnx_tflite: - inference_time: 20888.0 - throughput: 47.874377633090766 + inference_time: 14156.0 + throughput: 70.64142413111048 estimated_peak_memory_range: - min: 2527232 - max: 9840936 + min: 57344 + max: 2487920 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 787 - layers_on_gpu: 1 + layers_on_npu: 788 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 788 - job_id: jz5wyrm3g + job_id: jegnwxykg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,21 +211,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:50:31Z' + timestamp: '2024-08-27T01:02:54Z' - torchscript_onnx_tflite: - inference_time: 20939.0 - throughput: 47.75777257748699 + inference_time: 14101.0 + throughput: 70.916956244238 estimated_peak_memory_range: - min: 2543616 - max: 9702760 + min: 57344 + max: 4104000 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 787 - layers_on_gpu: 1 + layers_on_npu: 788 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 788 - job_id: jmg9oq9wg + job_id: jopr79q0g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -234,13 +234,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:50:32Z' + timestamp: '2024-08-27T01:02:55Z' - torchscript_onnx: - inference_time: 16978.0 - throughput: 58.89975262103899 + inference_time: 16871.0 + throughput: 59.273309228854245 estimated_peak_memory_range: - min: 84627456 - max: 84627456 + min: 84688896 + max: 84688896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -248,7 +248,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 737 - job_id: j2p0oxe9p + job_id: j1pvn4xkg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -257,4 +257,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:50:45Z' + timestamp: '2024-08-27T01:03:06Z' diff --git a/qai_hub_models/models/detr_resnet50_dc5/perf.yaml b/qai_hub_models/models/detr_resnet50_dc5/perf.yaml index 3c002119..91dcae88 100644 --- a/qai_hub_models/models/detr_resnet50_dc5/perf.yaml +++ b/qai_hub_models/models/detr_resnet50_dc5/perf.yaml @@ -45,26 +45,26 @@ models: - name: DETR-ResNet50-DC5 performance_metrics: - torchscript_onnx_tflite: - inference_time: 133136.0 - throughput: 7.511116452349477 + inference_time: 114929.0 + throughput: 8.70102411053781 estimated_peak_memory_range: - min: 9580544 - max: 17049688 + min: 176128 + max: 3388928 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 788 - layers_on_gpu: 1 + layers_on_npu: 789 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 789 - job_id: jlpe6nrog + job_id: jqp428q8g job_status: Passed torchscript_onnx: - inference_time: 116265.0 - throughput: 8.601040725927838 + inference_time: 116883.0 + throughput: 8.555564111119667 estimated_peak_memory_range: - min: 155648 - max: 100974512 + min: 139264 + max: 100582880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 737 - job_id: j1p8jk3k5 + job_id: j7gj81mv5 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,28 +81,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:49:56Z' + timestamp: '2024-08-27T01:02:21Z' - torchscript_onnx_tflite: - inference_time: 101299.0 - throughput: 9.871765762741981 + inference_time: 87231.0 + throughput: 11.463814469626623 estimated_peak_memory_range: - min: 9138176 - max: 454087584 + min: 192512 + max: 448623632 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 788 - layers_on_gpu: 1 + layers_on_npu: 789 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 789 - job_id: jygzz0xog + job_id: j0pxzmv35 job_status: Passed torchscript_onnx: - inference_time: 89431.0 - throughput: 11.181804966957767 + inference_time: 89365.0 + throughput: 11.190063223857214 estimated_peak_memory_range: - min: 1142784 - max: 406139152 + min: 0 + max: 405364128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 737 - job_id: jogk6klw5 + job_id: jlpen21op job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,21 +119,21 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:49:57Z' + timestamp: '2024-08-27T01:02:22Z' - torchscript_onnx_tflite: - inference_time: 132054.0 - throughput: 7.572659669529132 + inference_time: 113884.0 + throughput: 8.780864739559552 estimated_peak_memory_range: - min: 4071424 - max: 10093312 + min: 159744 + max: 3387944 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 788 - layers_on_gpu: 1 + layers_on_npu: 789 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 789 - job_id: jz5wyrd3g + job_id: jo5ml4rdg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,13 +142,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:49:41Z' + timestamp: '2024-08-27T01:02:08Z' - torchscript_onnx_tflite: - inference_time: 159267.0 - throughput: 6.278764590279216 + inference_time: 123415.0 + throughput: 8.102742778430498 estimated_peak_memory_range: - min: 9510912 - max: 413314128 + min: 225280 + max: 408732880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -156,7 +156,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 789 - job_id: jmg9oq3wg + job_id: jegnwx2kg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,21 +165,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:49:42Z' + timestamp: '2024-08-27T01:02:10Z' - torchscript_onnx_tflite: - inference_time: 142270.0 - throughput: 7.028888732691361 + inference_time: 119975.0 + throughput: 8.335069806209628 estimated_peak_memory_range: - min: 9555968 - max: 17428456 + min: 208896 + max: 2277800 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 788 - layers_on_gpu: 1 + layers_on_npu: 789 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 789 - job_id: jnp1omd85 + job_id: jopr79k0g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,21 +188,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:49:43Z' + timestamp: '2024-08-27T01:02:11Z' - torchscript_onnx_tflite: - inference_time: 142116.0 - throughput: 7.0365053899631285 + inference_time: 116631.0 + throughput: 8.574049780933029 estimated_peak_memory_range: - min: 2584576 - max: 11070264 + min: 184320 + max: 2362632 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 788 - layers_on_gpu: 1 + layers_on_npu: 789 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 789 - job_id: jvgd6mrrp + job_id: jep2zj8rp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,21 +211,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:49:44Z' + timestamp: '2024-08-27T01:02:12Z' - torchscript_onnx_tflite: - inference_time: 142539.0 - throughput: 7.01562379418966 + inference_time: 121019.0 + throughput: 8.263165288095259 estimated_peak_memory_range: - min: 9580544 - max: 16802848 + min: 61440 + max: 2052632 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 788 - layers_on_gpu: 1 + layers_on_npu: 789 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 789 - job_id: jz57o8jvg + job_id: jqpyyne8p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -234,13 +234,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:49:45Z' + timestamp: '2024-08-27T01:02:13Z' - torchscript_onnx: - inference_time: 117155.0 - throughput: 8.535700567624088 + inference_time: 117747.0 + throughput: 8.492785378820692 estimated_peak_memory_range: - min: 88743936 - max: 88743936 + min: 88875008 + max: 88875008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -248,7 +248,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 737 - job_id: jn5q4d7ng + job_id: jygz0w9o5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -257,4 +257,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:49:58Z' + timestamp: '2024-08-27T01:02:23Z' diff --git a/qai_hub_models/models/efficientnet_b0/perf.yaml b/qai_hub_models/models/efficientnet_b0/perf.yaml index 00df2237..1e4ca162 100644 --- a/qai_hub_models/models/efficientnet_b0/perf.yaml +++ b/qai_hub_models/models/efficientnet_b0/perf.yaml @@ -45,11 +45,11 @@ models: - name: EfficientNet-B0 performance_metrics: - torchscript_onnx_tflite: - inference_time: 1641.0 - throughput: 609.3845216331505 + inference_time: 1592.0 + throughput: 628.1407035175879 estimated_peak_memory_range: - min: 20480 - max: 1469776 + min: 24576 + max: 1447448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jz5wyr93g + job_id: jqpyyn97p job_status: Passed torchscript_onnx_qnn: - inference_time: 1690.0 - throughput: 591.7159763313609 + inference_time: 1669.0 + throughput: 599.1611743559017 estimated_peak_memory_range: - min: 16384 - max: 76702960 + min: 12288 + max: 88073712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 243 - job_id: jo5m9lzdg + job_id: j1p3r8ylp job_status: Passed torchscript_onnx: - inference_time: 1592.0 - throughput: 628.1407035175879 + inference_time: 1615.0 + throughput: 619.1950464396285 estimated_peak_memory_range: - min: 24576 - max: 16134704 + min: 12288 + max: 15676096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jn5q4d2ng + job_id: jnp1m302p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:49:08Z' + timestamp: '2024-08-27T01:01:36Z' - torchscript_onnx_tflite: - inference_time: 1133.0 - throughput: 882.61253309797 + inference_time: 1124.0 + throughput: 889.6797153024911 estimated_peak_memory_range: min: 16384 - max: 77138576 + max: 77651184 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jmg9oq4wg + job_id: j2p0xkn6p job_status: Passed torchscript_onnx_qnn: - inference_time: 1177.0 - throughput: 849.6176720475786 + inference_time: 1170.0 + throughput: 854.7008547008547 estimated_peak_memory_range: min: 0 - max: 16262032 + max: 19458224 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 243 - job_id: jegn1wekp + job_id: jwgo9mjxg job_status: Passed torchscript_onnx: - inference_time: 1172.0 - throughput: 853.2423208191126 + inference_time: 1174.0 + throughput: 851.7887563884157 estimated_peak_memory_range: - min: 0 - max: 82317296 + min: 466944 + max: 82722176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: j1glwqkjp + job_id: jvgdm0weg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:49:09Z' + timestamp: '2024-08-27T01:01:37Z' - torchscript_onnx_tflite: - inference_time: 1634.0 - throughput: 611.9951040391677 + inference_time: 1590.0 + throughput: 628.930817610063 estimated_peak_memory_range: - min: 12288 - max: 1467440 + min: 28672 + max: 1444104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jnp1om885 + job_id: j1p8k8lxp job_status: Passed torchscript_onnx_qnn: - inference_time: 1565.0 - throughput: 638.9776357827476 + inference_time: 1561.0 + throughput: 640.6149903907751 estimated_peak_memory_range: min: 638976 - max: 1982248 + max: 2371080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 243 - job_id: jep2ozmrg + job_id: j7gj81jx5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:49:02Z' + timestamp: '2024-08-27T01:01:31Z' - torchscript_onnx_tflite: - inference_time: 3008.0 - throughput: 332.4468085106383 + inference_time: 3002.0 + throughput: 333.11125916055965 estimated_peak_memory_range: min: 16384 - max: 84813232 + max: 84640912 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jvgd6mvrp + job_id: jogkkdj2g job_status: Passed torchscript_onnx_qnn: - inference_time: 3059.0 - throughput: 326.90421706440014 + inference_time: 3101.0 + throughput: 322.4766204450177 estimated_peak_memory_range: min: 618496 - max: 25723232 + max: 24923024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 243 - job_id: jogk6kyw5 + job_id: jmg9q8vlp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:49:07Z' + timestamp: '2024-08-27T01:01:35Z' - torchscript_onnx_tflite: - inference_time: 1646.0 - throughput: 607.5334143377886 + inference_time: 1593.0 + throughput: 627.7463904582548 estimated_peak_memory_range: - min: 36864 - max: 292933752 + min: 32768 + max: 1415536 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jz57o8dvg + job_id: jn5qdwj4g job_status: Passed torchscript_onnx_qnn: - inference_time: 1579.0 - throughput: 633.3122229259025 + inference_time: 1566.0 + throughput: 638.5696040868455 estimated_peak_memory_range: - min: 647168 - max: 1883752 + min: 634880 + max: 2270392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 243 - job_id: jqpy8yd8g + job_id: jlpen2j1p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:49:03Z' + timestamp: '2024-08-27T01:01:32Z' - torchscript_onnx_tflite: - inference_time: 1630.0 - throughput: 613.4969325153374 + inference_time: 1592.0 + throughput: 628.1407035175879 estimated_peak_memory_range: - min: 28672 - max: 17837184 + min: 24576 + max: 270089240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jqp4e2w8g + job_id: j1glq7j8p job_status: Passed torchscript_onnx_qnn: - inference_time: 1577.0 - throughput: 634.1154090044388 + inference_time: 1565.0 + throughput: 638.9776357827476 estimated_peak_memory_range: - min: 634880 - max: 1929200 + min: 638976 + max: 1934224 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 243 - job_id: j2p0oxr9p + job_id: jygz0w1k5 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:49:04Z' + timestamp: '2024-08-27T01:01:33Z' - torchscript_onnx_tflite: - inference_time: 1641.0 - throughput: 609.3845216331505 + inference_time: 1590.0 + throughput: 628.930817610063 estimated_peak_memory_range: - min: 12288 - max: 2187480 + min: 16384 + max: 4204592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: j0px0z13p + job_id: jw560vk05 job_status: Passed torchscript_onnx_qnn: - inference_time: 1583.0 - throughput: 631.7119393556538 + inference_time: 1574.0 + throughput: 635.3240152477764 estimated_peak_memory_range: - min: 630784 - max: 1973184 + min: 638976 + max: 1959112 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 243 - job_id: j1p8jk7k5 + job_id: jz5wrxo6p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:49:06Z' + timestamp: '2024-08-27T01:01:34Z' - torchscript_onnx_qnn: - inference_time: 1755.0 - throughput: 569.8005698005697 + inference_time: 1770.0 + throughput: 564.9717514124294 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 243 - job_id: joprx7y0p + job_id: j1pvn4jjg job_status: Passed torchscript_onnx: - inference_time: 1697.0 - throughput: 589.2751915144372 + inference_time: 1755.0 + throughput: 569.8005698005697 estimated_peak_memory_range: - min: 16076800 - max: 16076800 + min: 15831040 + max: 15831040 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jw56o0165 + job_id: jz5wrxo3p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:49:10Z' + timestamp: '2024-08-27T01:01:38Z' diff --git a/qai_hub_models/models/esrgan/perf.yaml b/qai_hub_models/models/esrgan/perf.yaml index 6367bd8d..56da4b84 100644 --- a/qai_hub_models/models/esrgan/perf.yaml +++ b/qai_hub_models/models/esrgan/perf.yaml @@ -45,11 +45,11 @@ models: - name: ESRGAN performance_metrics: - torchscript_onnx_tflite: - inference_time: 63878.0 - throughput: 15.65484204264379 + inference_time: 70939.0 + throughput: 14.096618221288713 estimated_peak_memory_range: - min: 3280896 - max: 5961080 + min: 3162112 + max: 5576896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1024 - job_id: j0px0zy1p + job_id: jvgdmzweg job_status: Passed torchscript_onnx_qnn: - inference_time: 64833.0 - throughput: 15.424243826446409 + inference_time: 70434.0 + throughput: 14.197688616293268 estimated_peak_memory_range: min: 102400 - max: 105054568 + max: 108968640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1026 - job_id: j1p8jkzx5 + job_id: jep2zv84p job_status: Passed torchscript_onnx: - inference_time: 69570.0 - throughput: 14.374011786689666 + inference_time: 68293.0 + throughput: 14.642789158478907 estimated_peak_memory_range: - min: 135168 - max: 44164912 + min: 139264 + max: 44386096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: j7gj38exp + job_id: j1p3rwklp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:48:27Z' + timestamp: '2024-08-27T11:56:09Z' - torchscript_onnx_tflite: - inference_time: 50620.0 - throughput: 19.755037534571315 + inference_time: 51258.0 + throughput: 19.509149791252096 estimated_peak_memory_range: min: 3260416 - max: 609041408 + max: 612109120 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1024 - job_id: jo5m9l3wg + job_id: jz5787zlp job_status: Passed torchscript_onnx_qnn: - inference_time: 49687.0 - throughput: 20.125988689194358 + inference_time: 49934.0 + throughput: 20.02643489406016 estimated_peak_memory_range: - min: 53248 - max: 98811568 + min: 73728 + max: 104441296 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1026 - job_id: jogk6k325 + job_id: jqpyy7e7p job_status: Passed torchscript_onnx: - inference_time: 50237.0 - throughput: 19.90564723211975 + inference_time: 51166.0 + throughput: 19.544228589297582 estimated_peak_memory_range: - min: 7294976 - max: 634344464 + min: 6709248 + max: 634927232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: jlpe6nk1g + job_id: jwgo94yxg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:48:28Z' + timestamp: '2024-08-27T11:56:10Z' - torchscript_onnx_tflite: - inference_time: 65091.0 - throughput: 15.363107034766712 + inference_time: 64927.0 + throughput: 15.401912917584363 estimated_peak_memory_range: - min: 3309568 - max: 5543048 + min: 3280896 + max: 6272432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1024 - job_id: jegn1w3rp + job_id: jqp429qvg job_status: Passed torchscript_onnx_qnn: - inference_time: 62657.0 - throughput: 15.959908709322182 + inference_time: 64742.0 + throughput: 15.445923820703717 estimated_peak_memory_range: - min: 360448 - max: 2060152 + min: 442368 + max: 1815472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1026 - job_id: j1glwq38p + job_id: j1p8k4oxp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:48:21Z' + timestamp: '2024-08-27T11:56:04Z' - torchscript_onnx_tflite: - inference_time: 156126.0 - throughput: 6.405083073927469 + inference_time: 133089.0 + throughput: 7.51376898165889 estimated_peak_memory_range: - min: 3211264 - max: 571696464 + min: 16384 + max: 571196848 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1024 - job_id: joprx7e9p + job_id: j0pxzdv15 job_status: Passed torchscript_onnx_qnn: - inference_time: 164307.0 - throughput: 6.0861679660635275 + inference_time: 141292.0 + throughput: 7.07754154516887 estimated_peak_memory_range: - min: 327680 - max: 77111888 + min: 335872 + max: 80302848 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1026 - job_id: j1pv2nvjg + job_id: jw560d605 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:48:26Z' + timestamp: '2024-08-27T11:56:08Z' - torchscript_onnx_tflite: - inference_time: 62949.0 - throughput: 15.88587586776597 + inference_time: 62292.0 + throughput: 16.05342580106595 estimated_peak_memory_range: - min: 3244032 - max: 6134584 + min: 3190784 + max: 5731456 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1024 - job_id: jep2ozl4g + job_id: jo5mldrwg job_status: Passed torchscript_onnx_qnn: - inference_time: 64830.0 - throughput: 15.424957581366652 + inference_time: 62758.0 + throughput: 15.934223525287612 estimated_peak_memory_range: - min: 393216 - max: 5202248 + min: 368640 + max: 1981024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1026 - job_id: jw56o0n05 + job_id: jogkk9z2g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:48:22Z' + timestamp: '2024-08-27T11:56:05Z' - torchscript_onnx_tflite: - inference_time: 63407.0 - throughput: 15.771129370574227 + inference_time: 63656.0 + throughput: 15.709438230488878 estimated_peak_memory_range: - min: 3256320 - max: 5566520 + min: 135168 + max: 2817616 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1024 - job_id: jqpy8y67g + job_id: jegnw72rg job_status: Passed torchscript_onnx_qnn: - inference_time: 64146.0 - throughput: 15.589436597761356 + inference_time: 64156.0 + throughput: 15.587006671238855 estimated_peak_memory_range: - min: 344064 - max: 1608320 + min: 401408 + max: 1728064 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1026 - job_id: j1p3orelp + job_id: jn5qdm84g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:48:23Z' + timestamp: '2024-08-27T11:56:06Z' - torchscript_onnx_tflite: - inference_time: 66683.0 - throughput: 14.996325900154462 + inference_time: 64701.0 + throughput: 15.455711658243304 estimated_peak_memory_range: - min: 3256320 - max: 6117632 + min: 3252224 + max: 5420872 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1024 - job_id: j2p0oxl6p + job_id: jopr7nk9g job_status: Passed torchscript_onnx_qnn: - inference_time: 64164.0 - throughput: 15.585063275356898 + inference_time: 64033.0 + throughput: 15.616947511439413 estimated_peak_memory_range: - min: 454656 - max: 1744104 + min: 389120 + max: 2067736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1026 - job_id: jwgod93x5 + job_id: j1glq1n8p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:48:24Z' + timestamp: '2024-08-27T11:56:07Z' - torchscript_onnx_qnn: - inference_time: 65373.0 - throughput: 15.29683508482095 + inference_time: 65253.0 + throughput: 15.324965901950868 estimated_peak_memory_range: - min: 221184 - max: 221184 + min: 208896 + max: 208896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1026 - job_id: jn5q4d34g + job_id: j2p0xvy6p job_status: Passed torchscript_onnx: - inference_time: 65982.0 - throughput: 15.155648510199752 + inference_time: 65443.0 + throughput: 15.280473083446664 estimated_peak_memory_range: - min: 41009152 - max: 41009152 + min: 40865792 + max: 40865792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: jygzz0rkg + job_id: j1pvn93jg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:48:29Z' + timestamp: '2024-08-27T11:56:11Z' diff --git a/qai_hub_models/models/fastsam_s/perf.yaml b/qai_hub_models/models/fastsam_s/perf.yaml index bd10ec6c..ae5873ee 100644 --- a/qai_hub_models/models/fastsam_s/perf.yaml +++ b/qai_hub_models/models/fastsam_s/perf.yaml @@ -21,7 +21,6 @@ aggregated: - QCS8550 (Proxy) - SA8775 (Proxy) - SA8650 (Proxy) - - SA8255 (Proxy) - QCS8450 (Proxy) - XR2 Gen 2 (Proxy) - Google Pixel 5a 5G @@ -39,17 +38,16 @@ aggregated: - Qcs8550 - Sa8775p - Sa8650p - - Sa8255p - Qcs8450 models: - name: FastSam-S performance_metrics: - torchscript_onnx_qnn: - inference_time: 8081.0 - throughput: 123.74706100730107 + inference_time: 8026.0 + throughput: 124.595066035385 estimated_peak_memory_range: - min: 6172672 - max: 24265104 + min: 4214784 + max: 21921520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +55,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 286 - job_id: j1glwq28p + job_id: jygz0wqk5 job_status: Passed torchscript_onnx: - inference_time: 10524.0 - throughput: 95.02090459901179 + inference_time: 10531.0 + throughput: 94.95774380400722 estimated_peak_memory_range: - min: 2129920 - max: 28908928 + min: 2138112 + max: 28184024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +70,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: jz5wyrq6g + job_id: jo5ml47wg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,13 +79,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:47:34Z' + timestamp: '2024-08-27T01:00:11Z' - torchscript_onnx_qnn: - inference_time: 6029.0 - throughput: 165.8649859014762 + inference_time: 5982.0 + throughput: 167.16817118020728 estimated_peak_memory_range: min: 4931584 - max: 37170720 + max: 37961888 primary_compute_unit: NPU precision: fp16 layer_info: @@ -95,14 +93,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 286 - job_id: jw56o0z05 + job_id: jz5wrxk6p job_status: Passed torchscript_onnx: - inference_time: 7466.0 - throughput: 133.9405304045004 + inference_time: 7354.0 + throughput: 135.98041881968996 estimated_peak_memory_range: - min: 16449536 - max: 88110624 + min: 15728640 + max: 89275104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +108,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: jmg9oqwlg + job_id: jegnwx4rg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,13 +117,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:47:35Z' + timestamp: '2024-08-27T01:00:12Z' - torchscript_onnx_qnn: - inference_time: 7777.0 - throughput: 128.5842870001286 + inference_time: 7866.0 + throughput: 127.12941774726671 estimated_peak_memory_range: - min: 4997120 - max: 6469320 + min: 4993024 + max: 10330720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -133,7 +131,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 286 - job_id: jwgod9nx5 + job_id: jnp1m392p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,13 +140,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:47:28Z' + timestamp: '2024-08-27T01:00:06Z' - torchscript_onnx_qnn: - inference_time: 12754.0 - throughput: 78.40677434530343 + inference_time: 12997.0 + throughput: 76.94083249980764 estimated_peak_memory_range: - min: 4947968 - max: 40506512 + min: 4931584 + max: 42015344 primary_compute_unit: NPU precision: fp16 layer_info: @@ -156,7 +154,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 286 - job_id: jygzz0jkg + job_id: j0pxzmq15 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,13 +163,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:47:33Z' + timestamp: '2024-08-27T01:00:10Z' - torchscript_onnx_qnn: - inference_time: 7776.0 - throughput: 128.6008230452675 + inference_time: 7921.0 + throughput: 126.24668602449185 estimated_peak_memory_range: - min: 4997120 - max: 6592904 + min: 4988928 + max: 10198152 primary_compute_unit: NPU precision: fp16 layer_info: @@ -179,7 +177,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 286 - job_id: j1pv2nrjg + job_id: jvgdm0keg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,13 +186,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:47:29Z' + timestamp: '2024-08-27T01:00:07Z' - torchscript_onnx_qnn: - inference_time: 7871.0 - throughput: 127.04865963664083 + inference_time: 7914.0 + throughput: 126.35835228708618 estimated_peak_memory_range: min: 4988928 - max: 10071552 + max: 6161120 primary_compute_unit: NPU precision: fp16 layer_info: @@ -202,7 +200,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 286 - job_id: j7gj382xp + job_id: jz5786mlp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,36 +209,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:47:31Z' - - torchscript_onnx_qnn: - inference_time: 7811.0 - throughput: 128.02458071949815 - estimated_peak_memory_range: - min: 5001216 - max: 6277216 - primary_compute_unit: NPU - precision: fp16 - layer_info: - layers_on_npu: 286 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 286 - job_id: jlpe6nw1g - job_status: Passed - reference_device_info: - name: SA8255 (Proxy) - os: '13' - form_factor: Auto - os_name: Android - manufacturer: Qualcomm - chipset: Sa8255p - timestamp: '2024-08-10T23:47:32Z' + timestamp: '2024-08-27T01:00:08Z' - torchscript_onnx_qnn: - inference_time: 8182.0 - throughput: 122.21950623319482 + inference_time: 8309.0 + throughput: 120.35142616440005 estimated_peak_memory_range: - min: 4923392 - max: 4923392 + min: 4927488 + max: 4927488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -248,14 +223,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 286 - job_id: j1p3or1lp + job_id: jmg9q8rlp job_status: Passed torchscript_onnx: - inference_time: 10622.0 - throughput: 94.14422895876483 + inference_time: 10861.0 + throughput: 92.07255317189946 estimated_peak_memory_range: - min: 22618112 - max: 22618112 + min: 23613440 + max: 23613440 primary_compute_unit: NPU precision: fp16 layer_info: @@ -263,7 +238,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: jnp1ome25 + job_id: jopr79r9g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -272,4 +247,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:47:36Z' + timestamp: '2024-08-27T01:00:13Z' diff --git a/qai_hub_models/models/fastsam_x/perf.yaml b/qai_hub_models/models/fastsam_x/perf.yaml index 3e89f58c..0d828436 100644 --- a/qai_hub_models/models/fastsam_x/perf.yaml +++ b/qai_hub_models/models/fastsam_x/perf.yaml @@ -45,11 +45,11 @@ models: - name: FastSam-X performance_metrics: - torchscript_onnx_qnn: - inference_time: 45669.0 - throughput: 21.89669140992796 + inference_time: 45740.0 + throughput: 21.862702229995627 estimated_peak_memory_range: - min: 5349376 - max: 23516072 + min: 5103616 + max: 21629672 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 418 - job_id: jwgod9kx5 + job_id: jlpen2n1p job_status: Passed torchscript_onnx: - inference_time: 48725.0 - throughput: 20.52334530528476 + inference_time: 50890.0 + throughput: 19.650225977598744 estimated_peak_memory_range: - min: 24576 - max: 214872088 + min: 118784 + max: 165544320 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 421 - job_id: jvgd6mqep + job_id: j0pxzm815 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,13 +81,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:46:53Z' + timestamp: '2024-08-27T00:59:33Z' - torchscript_onnx_qnn: - inference_time: 34455.0 - throughput: 29.02336380786533 + inference_time: 34391.0 + throughput: 29.077374894594517 estimated_peak_memory_range: - min: 4931584 - max: 59121856 + min: 4956160 + max: 57199632 primary_compute_unit: NPU precision: fp16 layer_info: @@ -95,14 +95,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 418 - job_id: j1pv2n0jg + job_id: jygz0w0k5 job_status: Passed torchscript_onnx: - inference_time: 36164.0 - throughput: 27.65180842827121 + inference_time: 36136.0 + throughput: 27.67323444764224 estimated_peak_memory_range: - min: 3309568 - max: 142076208 + min: 0 + max: 139940448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 421 - job_id: jz57o8llg + job_id: jo5ml41wg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,13 +119,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:46:54Z' + timestamp: '2024-08-27T00:59:34Z' - torchscript_onnx_qnn: - inference_time: 42982.0 - throughput: 23.265553022195338 + inference_time: 42639.0 + throughput: 23.452707615094162 estimated_peak_memory_range: - min: 5033984 - max: 10209720 + min: 5095424 + max: 10176336 primary_compute_unit: NPU precision: fp16 layer_info: @@ -133,7 +133,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 418 - job_id: jlpe6ne1g + job_id: jmg9q87lp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,13 +142,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:46:46Z' + timestamp: '2024-08-27T00:59:27Z' - torchscript_onnx_qnn: - inference_time: 89049.0 - throughput: 11.22977237251401 + inference_time: 90144.0 + throughput: 11.093361732339368 estimated_peak_memory_range: - min: 0 - max: 53289296 + min: 4931584 + max: 52517392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -156,7 +156,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 418 - job_id: jnp1omw25 + job_id: jqp4286vg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,13 +165,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:46:51Z' + timestamp: '2024-08-27T00:59:31Z' - torchscript_onnx_qnn: - inference_time: 44117.0 - throughput: 22.666999115987036 + inference_time: 43081.0 + throughput: 23.21208885587614 estimated_peak_memory_range: - min: 5095424 - max: 6231968 + min: 5042176 + max: 10295768 primary_compute_unit: NPU precision: fp16 layer_info: @@ -179,7 +179,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 418 - job_id: jygzz0okg + job_id: jnp1m3k2p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,13 +188,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:46:47Z' + timestamp: '2024-08-27T00:59:28Z' - torchscript_onnx_qnn: - inference_time: 43149.0 - throughput: 23.175508123015597 + inference_time: 42771.0 + throughput: 23.380327792195647 estimated_peak_memory_range: - min: 5070848 - max: 6409216 + min: 5066752 + max: 6349336 primary_compute_unit: NPU precision: fp16 layer_info: @@ -202,7 +202,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 418 - job_id: jz5wyr36g + job_id: jvgdm0yeg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,13 +211,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:46:49Z' + timestamp: '2024-08-27T00:59:29Z' - torchscript_onnx_qnn: - inference_time: 43251.0 - throughput: 23.120852697047468 + inference_time: 43641.0 + throughput: 22.91423202951353 estimated_peak_memory_range: - min: 5066752 - max: 6336032 + min: 5070848 + max: 6402896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -225,7 +225,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 418 - job_id: jmg9oqylg + job_id: jz57861lp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -234,10 +234,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:46:50Z' + timestamp: '2024-08-27T00:59:30Z' - torchscript_onnx_qnn: - inference_time: 44539.0 - throughput: 22.452232874559375 + inference_time: 44291.0 + throughput: 22.57795037366508 estimated_peak_memory_range: min: 4923392 max: 4923392 @@ -248,14 +248,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 418 - job_id: j7gj38zxp + job_id: jz5wrx06p job_status: Passed torchscript_onnx: - inference_time: 49433.0 - throughput: 20.22940141201222 + inference_time: 49381.0 + throughput: 20.25070371195399 estimated_peak_memory_range: - min: 147279872 - max: 147279872 + min: 147308544 + max: 147308544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -263,7 +263,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 421 - job_id: jqp4e2dvg + job_id: jegnwxdrg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -272,4 +272,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:46:55Z' + timestamp: '2024-08-27T00:59:35Z' diff --git a/qai_hub_models/models/fcn_resnet50/perf.yaml b/qai_hub_models/models/fcn_resnet50/perf.yaml index eccd6efb..fa53943e 100644 --- a/qai_hub_models/models/fcn_resnet50/perf.yaml +++ b/qai_hub_models/models/fcn_resnet50/perf.yaml @@ -45,11 +45,11 @@ models: - name: FCN-ResNet50 performance_metrics: - torchscript_onnx_tflite: - inference_time: 41772.0 - throughput: 23.939480992052093 + inference_time: 41394.0 + throughput: 24.15809054452336 estimated_peak_memory_range: - min: 21966848 - max: 23833760 + min: 22102016 + max: 24365520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jogk6kv25 + job_id: jw560vm75 job_status: Passed torchscript_onnx_qnn: - inference_time: 42966.0 - throughput: 23.27421682260392 + inference_time: 42319.0 + throughput: 23.630047968997378 estimated_peak_memory_range: - min: 3219456 - max: 20703168 + min: 3203072 + max: 20113384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 127 - job_id: j7gj38dxp + job_id: jmg9q82vp job_status: Passed torchscript_onnx: - inference_time: 42911.0 - throughput: 23.30404791312251 + inference_time: 43418.0 + throughput: 23.031922244230504 estimated_peak_memory_range: - min: 22216704 - max: 104927800 + min: 23023616 + max: 105749480 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: jqp4e2kvg + job_id: jqp4282vg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:46:05Z' + timestamp: '2024-08-27T00:58:48Z' - torchscript_onnx_tflite: - inference_time: 31017.0 - throughput: 32.24038430538092 + inference_time: 30751.0 + throughput: 32.51926766609216 estimated_peak_memory_range: - min: 20045824 - max: 159706496 + min: 20713472 + max: 160136784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jn5q4d04g + job_id: j1p3r87zp job_status: Passed torchscript_onnx_qnn: - inference_time: 31331.0 - throughput: 31.917270435032396 + inference_time: 31019.0 + throughput: 32.238305554660045 estimated_peak_memory_range: min: 3162112 - max: 62858128 + max: 57481920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 127 - job_id: jlpe6no1g + job_id: jnp1m31lp job_status: Passed torchscript_onnx: - inference_time: 31886.0 - throughput: 31.361726149407264 + inference_time: 31759.0 + throughput: 31.487137504329482 estimated_peak_memory_range: - min: 782336 - max: 139830384 + min: 51134464 + max: 192504320 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: j0px0zn1p + job_id: j0pxzmz15 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:46:06Z' + timestamp: '2024-08-27T00:58:49Z' - torchscript_onnx_tflite: - inference_time: 41899.0 - throughput: 23.866918064870283 + inference_time: 41340.0 + throughput: 24.189646831156264 estimated_peak_memory_range: - min: 22122496 - max: 108355328 + min: 22020096 + max: 24569544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j1glwq48p + job_id: jwgo9mwdg job_status: Passed torchscript_onnx_qnn: - inference_time: 39385.0 - throughput: 25.39037704709915 + inference_time: 38892.0 + throughput: 25.712228735986834 estimated_peak_memory_range: - min: 3293184 - max: 4863056 + min: 3284992 + max: 4643776 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 127 - job_id: jz5wyr26g + job_id: jz5wrxr6p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:45:59Z' + timestamp: '2024-08-27T00:58:43Z' - torchscript_onnx_tflite: - inference_time: 66730.0 - throughput: 14.985763524651581 + inference_time: 66783.0 + throughput: 14.973870595810311 estimated_peak_memory_range: - min: 22142976 - max: 108550784 + min: 22179840 + max: 108633488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jw56o0205 + job_id: j1pvn4mmg job_status: Passed torchscript_onnx_qnn: - inference_time: 66095.0 - throughput: 15.129737499054391 + inference_time: 65705.0 + throughput: 15.219541891789056 estimated_peak_memory_range: - min: 3203072 - max: 39398400 + min: 3182592 + max: 38054816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 127 - job_id: jz57o80lg + job_id: jz57868lp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:46:03Z' + timestamp: '2024-08-27T00:58:47Z' - torchscript_onnx_tflite: - inference_time: 41501.0 - throughput: 24.095804920363364 + inference_time: 41575.0 + throughput: 24.052916416115455 estimated_peak_memory_range: - min: 22122496 - max: 24302976 + min: 22110208 + max: 24032392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j1p3ornlp + job_id: j7gj81y85 job_status: Passed torchscript_onnx_qnn: - inference_time: 39969.0 - throughput: 25.019390027271136 + inference_time: 39271.0 + throughput: 25.464082911053957 estimated_peak_memory_range: - min: 3342336 - max: 4580624 + min: 3284992 + max: 4673960 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 127 - job_id: jmg9oqjlg + job_id: jmg9q8qlp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:46:00Z' + timestamp: '2024-08-27T00:58:44Z' - torchscript_onnx_tflite: - inference_time: 42319.0 - throughput: 23.630047968997378 + inference_time: 41308.0 + throughput: 24.208385784835865 estimated_peak_memory_range: - min: 22118400 - max: 24156040 + min: 22097920 + max: 24331576 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jwgod9zx5 + job_id: jlpen2x0p job_status: Passed torchscript_onnx_qnn: - inference_time: 40177.0 - throughput: 24.889862359061155 + inference_time: 39020.0 + throughput: 25.627883136852898 estimated_peak_memory_range: - min: 3309568 - max: 4532256 + min: 3289088 + max: 4812920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 127 - job_id: jnp1omy25 + job_id: jnp1m3m2p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:46:01Z' + timestamp: '2024-08-27T00:58:45Z' - torchscript_onnx_tflite: - inference_time: 41668.0 - throughput: 23.999232024575214 + inference_time: 41894.0 + throughput: 23.869766553683107 estimated_peak_memory_range: - min: 22114304 - max: 24211872 + min: 22085632 + max: 24208352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j1pv2nqjg + job_id: jygz0wy65 job_status: Passed torchscript_onnx_qnn: - inference_time: 39795.0 - throughput: 25.128785023244127 + inference_time: 39162.0 + throughput: 25.534957356621213 estimated_peak_memory_range: - min: 3293184 - max: 4575808 + min: 3256320 + max: 4496128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 127 - job_id: jvgd6meep + job_id: jvgdm0meg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:46:02Z' + timestamp: '2024-08-27T00:58:46Z' - torchscript_onnx_qnn: - inference_time: 39586.0 - throughput: 25.261456070327895 + inference_time: 39171.0 + throughput: 25.5290903985091 estimated_peak_memory_range: min: 3153920 max: 3153920 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 127 - job_id: jygzz02kg + job_id: jvgdm04lg job_status: Passed torchscript_onnx: - inference_time: 42300.0 - throughput: 23.64066193853428 + inference_time: 42041.0 + throughput: 23.786303846245332 estimated_peak_memory_range: - min: 69472256 - max: 69472256 + min: 69488640 + max: 69488640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: jo5m9lqwg + job_id: jo5ml4lwg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:46:07Z' + timestamp: '2024-08-27T00:58:50Z' diff --git a/qai_hub_models/models/fcn_resnet50_quantized/perf.yaml b/qai_hub_models/models/fcn_resnet50_quantized/perf.yaml index 094b9a08..f8397b92 100644 --- a/qai_hub_models/models/fcn_resnet50_quantized/perf.yaml +++ b/qai_hub_models/models/fcn_resnet50_quantized/perf.yaml @@ -48,11 +48,11 @@ models: - name: FCN-ResNet50-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 13313.0 - throughput: 75.11454968827462 + inference_time: 13269.0 + throughput: 75.36362951239732 estimated_peak_memory_range: - min: 5541888 - max: 53740408 + min: 5537792 + max: 316797728 primary_compute_unit: NPU precision: int8 layer_info: @@ -60,14 +60,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 89 - job_id: jw56o0y75 + job_id: jogkkd6og job_status: Passed torchscript_onnx_qnn: - inference_time: 14778.0 - throughput: 67.66815536608472 + inference_time: 14742.0 + throughput: 67.8334011667345 estimated_peak_memory_range: - min: 45056 - max: 12104736 + min: 12288 + max: 46423296 primary_compute_unit: NPU precision: int8 layer_info: @@ -75,14 +75,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jnp1om4l5 + job_id: jygz0wz65 job_status: Passed torchscript_onnx: - inference_time: 12663.0 - throughput: 78.97022822395957 + inference_time: 12764.0 + throughput: 78.34534628643058 estimated_peak_memory_range: min: 16384 - max: 43726344 + max: 43615328 primary_compute_unit: NPU precision: int8 layer_info: @@ -90,7 +90,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 80 - job_id: jo5m9lewg + job_id: jegnwxqmg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -99,13 +99,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:45:18Z' + timestamp: '2024-08-27T00:58:06Z' - torchscript_onnx_tflite: - inference_time: 9456.0 - throughput: 105.75296108291032 + inference_time: 9444.0 + throughput: 105.8873358746294 estimated_peak_memory_range: - min: 3620864 - max: 88292240 + min: 3514368 + max: 92911952 primary_compute_unit: NPU precision: int8 layer_info: @@ -113,14 +113,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 89 - job_id: j1p3orjzp + job_id: jn5qdw4mg job_status: Passed torchscript_onnx_qnn: - inference_time: 10871.0 - throughput: 91.98785760279642 + inference_time: 10885.0 + throughput: 91.86954524575103 estimated_peak_memory_range: min: 802816 - max: 32185088 + max: 33772848 primary_compute_unit: NPU precision: int8 layer_info: @@ -128,14 +128,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jvgd6mxlp + job_id: jz5wrxyjp job_status: Passed torchscript_onnx: - inference_time: 9289.0 - throughput: 107.65421466250403 + inference_time: 9257.0 + throughput: 108.02635843145728 estimated_peak_memory_range: - min: 139264 - max: 101471856 + min: 122880 + max: 104530672 primary_compute_unit: NPU precision: int8 layer_info: @@ -143,7 +143,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 80 - job_id: jegn1w0rp + job_id: jep2zjdmp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -152,13 +152,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:45:20Z' + timestamp: '2024-08-27T00:58:07Z' - torchscript_onnx_tflite: - inference_time: 13383.0 - throughput: 74.72166180975864 + inference_time: 13277.0 + throughput: 75.31821947729156 estimated_peak_memory_range: - min: 5570560 - max: 11750944 + min: 5550080 + max: 17081320 primary_compute_unit: NPU precision: int8 layer_info: @@ -166,14 +166,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 89 - job_id: jwgod92d5 + job_id: j1glq7wlp job_status: Passed torchscript_onnx_qnn: - inference_time: 12837.0 - throughput: 77.8998208304121 + inference_time: 12640.0 + throughput: 79.11392405063292 estimated_peak_memory_range: - min: 811008 - max: 2124408 + min: 868352 + max: 2181920 primary_compute_unit: NPU precision: int8 layer_info: @@ -181,7 +181,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jmg9oq0lg + job_id: jnp1m3olp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -190,13 +190,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:45:11Z' + timestamp: '2024-08-27T00:58:00Z' - torchscript_onnx_tflite: - inference_time: 15773.0 - throughput: 63.39948012426298 + inference_time: 15672.0 + throughput: 63.808065339458906 estimated_peak_memory_range: - min: 5615616 - max: 92128064 + min: 5636096 + max: 95800928 primary_compute_unit: NPU precision: int8 layer_info: @@ -204,14 +204,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 89 - job_id: j1pv2n6mg + job_id: jw560vo75 job_status: Passed torchscript_onnx_qnn: - inference_time: 17044.0 - throughput: 58.67167331612298 + inference_time: 17067.0 + throughput: 58.59260561317162 estimated_peak_memory_range: - min: 827392 - max: 32628272 + min: 819200 + max: 37857504 primary_compute_unit: NPU precision: int8 layer_info: @@ -219,7 +219,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jqp4e2nvg + job_id: j0pxzmr95 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -228,13 +228,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:45:16Z' + timestamp: '2024-08-27T00:58:04Z' - torchscript_onnx_tflite: - inference_time: 13328.0 - throughput: 75.03001200480192 + inference_time: 13295.0 + throughput: 75.21624670928921 estimated_peak_memory_range: - min: 5529600 - max: 30233256 + min: 5521408 + max: 35469680 primary_compute_unit: NPU precision: int8 layer_info: @@ -242,14 +242,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 89 - job_id: j7gj38v8p + job_id: j1p3r8ozp job_status: Passed torchscript_onnx_qnn: - inference_time: 12635.0 - throughput: 79.14523149980214 + inference_time: 12786.0 + throughput: 78.21054278116691 estimated_peak_memory_range: - min: 839680 - max: 2164224 + min: 2265088 + max: 3472496 primary_compute_unit: NPU precision: int8 layer_info: @@ -257,7 +257,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jnp1om225 + job_id: jvgdm06lg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -266,13 +266,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:45:12Z' + timestamp: '2024-08-27T00:58:01Z' - torchscript_onnx_tflite: - inference_time: 13360.0 - throughput: 74.8502994011976 + inference_time: 13233.0 + throughput: 75.56865412227008 estimated_peak_memory_range: - min: 5550080 - max: 30353384 + min: 5529600 + max: 35544080 primary_compute_unit: NPU precision: int8 layer_info: @@ -280,14 +280,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 89 - job_id: jlpe6nd0g + job_id: jwgo9mddg job_status: Passed torchscript_onnx_qnn: - inference_time: 12604.0 - throughput: 79.33989209774674 + inference_time: 12764.0 + throughput: 78.34534628643058 estimated_peak_memory_range: - min: 884736 - max: 2161064 + min: 819200 + max: 2473608 primary_compute_unit: NPU precision: int8 layer_info: @@ -295,7 +295,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jvgd6mnep + job_id: jz5786nrp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -304,13 +304,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:45:14Z' + timestamp: '2024-08-27T00:58:02Z' - torchscript_onnx_tflite: - inference_time: 13365.0 - throughput: 74.82229704451926 + inference_time: 13259.0 + throughput: 75.4204691153179 estimated_peak_memory_range: - min: 16384 - max: 14638536 + min: 5574656 + max: 172574016 primary_compute_unit: NPU precision: int8 layer_info: @@ -318,14 +318,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 89 - job_id: jygzz036g + job_id: j1pvn42mg job_status: Passed torchscript_onnx_qnn: - inference_time: 12503.0 - throughput: 79.98080460689434 + inference_time: 12284.0 + throughput: 81.406707912732 estimated_peak_memory_range: - min: 819200 - max: 2102704 + min: 856064 + max: 2109088 primary_compute_unit: NPU precision: int8 layer_info: @@ -333,7 +333,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jz57o82lg + job_id: jqp4284lg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -342,13 +342,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:45:15Z' + timestamp: '2024-08-27T00:58:03Z' - torchscript_onnx_tflite: - inference_time: 87707.0 - throughput: 11.401598504110277 + inference_time: 88241.0 + throughput: 11.332600491834862 estimated_peak_memory_range: - min: 6070272 - max: 94347568 + min: 5795840 + max: 100746880 primary_compute_unit: NPU precision: int8 layer_info: @@ -356,14 +356,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 89 - job_id: jz5wyrejg + job_id: j7gj81385 job_status: Passed torchscript_onnx_qnn: - inference_time: 112068.0 - throughput: 8.923153799478888 + inference_time: 116726.0 + throughput: 8.567071603584463 estimated_peak_memory_range: - min: 999424 - max: 8898032 + min: 1028096 + max: 9412640 primary_compute_unit: NPU precision: int8 layer_info: @@ -371,7 +371,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j0px0z91p + job_id: jo5ml4kqg job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -380,13 +380,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:45:17Z' + timestamp: '2024-08-27T00:58:05Z' - torchscript_onnx_qnn: - inference_time: 12718.0 - throughput: 78.62871520679352 + inference_time: 13011.0 + throughput: 76.85804319422027 estimated_peak_memory_range: - min: 798720 - max: 798720 + min: 794624 + max: 794624 primary_compute_unit: NPU precision: int8 layer_info: @@ -394,14 +394,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jz5wyrw6g + job_id: jmg9q8ovp job_status: Passed torchscript_onnx: - inference_time: 12585.0 - throughput: 79.45967421533572 + inference_time: 12611.0 + throughput: 79.29585282689715 estimated_peak_memory_range: - min: 35409920 - max: 35409920 + min: 35115008 + max: 35115008 primary_compute_unit: NPU precision: int8 layer_info: @@ -409,7 +409,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 80 - job_id: joprx769p + job_id: jqpyyn24p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -418,4 +418,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:45:21Z' + timestamp: '2024-08-27T00:58:08Z' diff --git a/qai_hub_models/models/ffnet_122ns_lowres/perf.yaml b/qai_hub_models/models/ffnet_122ns_lowres/perf.yaml index 5734b97a..d307b3f3 100644 --- a/qai_hub_models/models/ffnet_122ns_lowres/perf.yaml +++ b/qai_hub_models/models/ffnet_122ns_lowres/perf.yaml @@ -45,11 +45,11 @@ models: - name: FFNet-122NS-LowRes performance_metrics: - torchscript_onnx_tflite: - inference_time: 7042.0 - throughput: 142.00511218403864 + inference_time: 7060.0 + throughput: 141.643059490085 estimated_peak_memory_range: - min: 53248 - max: 2027520 + min: 0 + max: 2196520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 216 - job_id: jwgod96d5 + job_id: j1glq7ylp job_status: Passed torchscript_onnx_qnn: - inference_time: 6998.0 - throughput: 142.89797084881394 + inference_time: 7115.0 + throughput: 140.54813773717498 estimated_peak_memory_range: - min: 6307840 - max: 29654552 + min: 6316032 + max: 30330536 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 348 - job_id: jnp1omzl5 + job_id: jygz0wl65 job_status: Passed torchscript_onnx: - inference_time: 7875.0 - throughput: 126.98412698412699 + inference_time: 7809.0 + throughput: 128.05736970162633 estimated_peak_memory_range: - min: 6311936 - max: 9155112 + min: 49152 + max: 129267592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 350 - job_id: jep2ozkmg + job_id: jo5ml49qg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:43:40Z' + timestamp: '2024-08-27T00:56:32Z' - torchscript_onnx_tflite: - inference_time: 4973.0 - throughput: 201.08586366378444 + inference_time: 4936.0 + throughput: 202.5931928687196 estimated_peak_memory_range: - min: 659456 - max: 69693056 + min: 860160 + max: 70027168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 216 - job_id: j1pv2nkmg + job_id: jw560v875 job_status: Passed torchscript_onnx_qnn: - inference_time: 4933.0 - throughput: 202.71639975674032 + inference_time: 4925.0 + throughput: 203.0456852791878 estimated_peak_memory_range: min: 6307840 - max: 30325920 + max: 33045680 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 348 - job_id: jvgd6m1lp + job_id: jz5wrxljp job_status: Passed torchscript_onnx: - inference_time: 5698.0 - throughput: 175.5001755001755 + inference_time: 5558.0 + throughput: 179.92083483267362 estimated_peak_memory_range: - min: 0 - max: 81815680 + min: 933888 + max: 84211328 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 350 - job_id: jqpy8y14g + job_id: jegnwx1mg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:43:41Z' + timestamp: '2024-08-27T00:56:33Z' - torchscript_onnx_tflite: - inference_time: 7053.0 - throughput: 141.7836381681554 + inference_time: 6997.0 + throughput: 142.91839359725597 estimated_peak_memory_range: - min: 655360 - max: 2401104 + min: 638976 + max: 3108008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 216 - job_id: j7gj38n8p + job_id: j1p3r8zzp job_status: Passed torchscript_onnx_qnn: - inference_time: 6038.0 - throughput: 165.61775422325275 + inference_time: 6022.0 + throughput: 166.05778811026238 estimated_peak_memory_range: - min: 6377472 - max: 7571512 + min: 6332416 + max: 7962816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 348 - job_id: jqp4e2llg + job_id: jnp1m3nlp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:43:34Z' + timestamp: '2024-08-27T00:56:27Z' - torchscript_onnx_tflite: - inference_time: 9172.0 - throughput: 109.02747492368077 + inference_time: 9011.0 + throughput: 110.97547442015315 estimated_peak_memory_range: - min: 638976 - max: 62460896 + min: 663552 + max: 62695392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 216 - job_id: jlpe6nm0g + job_id: jwgo9mldg job_status: Passed torchscript_onnx_qnn: - inference_time: 9066.0 - throughput: 110.30222810500773 + inference_time: 9197.0 + throughput: 108.73110796999022 estimated_peak_memory_range: - min: 6307840 - max: 27568160 + min: 6324224 + max: 28577968 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 348 - job_id: joprx7vep + job_id: j0pxzm095 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:43:39Z' + timestamp: '2024-08-27T00:56:31Z' - torchscript_onnx_tflite: - inference_time: 7062.0 - throughput: 141.6029453412631 + inference_time: 7084.0 + throughput: 141.1631846414455 estimated_peak_memory_range: - min: 659456 - max: 3222352 + min: 643072 + max: 2920696 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 216 - job_id: jygzz0d6g + job_id: j1pvn4lmg job_status: Passed torchscript_onnx_qnn: - inference_time: 6141.0 - throughput: 162.83992835043153 + inference_time: 5984.0 + throughput: 167.11229946524065 estimated_peak_memory_range: - min: 6377472 - max: 7662456 + min: 6340608 + max: 7701504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 348 - job_id: j0px0zk9p + job_id: jvgdm0dlg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:43:35Z' + timestamp: '2024-08-27T00:56:28Z' - torchscript_onnx_tflite: - inference_time: 7147.0 - throughput: 139.91884706870016 + inference_time: 7029.0 + throughput: 142.2677479015507 estimated_peak_memory_range: - min: 708608 - max: 3132384 + min: 643072 + max: 2632776 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 216 - job_id: jz5wyr6jg + job_id: j7gj81r85 job_status: Passed torchscript_onnx_qnn: - inference_time: 6024.0 - throughput: 166.00265604249668 + inference_time: 6058.0 + throughput: 165.0709805216243 estimated_peak_memory_range: - min: 6377472 - max: 7699632 + min: 6361088 + max: 8142880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 348 - job_id: jo5m9lnqg + job_id: jz5786orp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:43:36Z' + timestamp: '2024-08-27T00:56:29Z' - torchscript_onnx_tflite: - inference_time: 7168.0 - throughput: 139.50892857142858 + inference_time: 7019.0 + throughput: 142.47043738424276 estimated_peak_memory_range: - min: 28672 - max: 77588520 + min: 643072 + max: 2711672 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 216 - job_id: jmg9oqnvg + job_id: jlpen270p job_status: Passed torchscript_onnx_qnn: - inference_time: 6029.0 - throughput: 165.8649859014762 + inference_time: 6106.0 + throughput: 163.77333770062233 estimated_peak_memory_range: - min: 6356992 - max: 7677744 + min: 6361088 + max: 7856264 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 348 - job_id: jegn1w6mp + job_id: jqp428elg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:43:38Z' + timestamp: '2024-08-27T00:56:30Z' - torchscript_onnx_qnn: - inference_time: 6314.0 - throughput: 158.37820715869498 + inference_time: 6431.0 + throughput: 155.49681231534754 estimated_peak_memory_range: min: 6303744 max: 6303744 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 348 - job_id: jz57o8yrg + job_id: jmg9q8zvp job_status: Passed torchscript_onnx: - inference_time: 7610.0 - throughput: 131.4060446780552 + inference_time: 7580.0 + throughput: 131.92612137203167 estimated_peak_memory_range: - min: 61136896 - max: 61136896 + min: 60731392 + max: 60731392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 350 - job_id: j2p0oxzep + job_id: jopr79xeg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:43:42Z' + timestamp: '2024-08-27T00:56:34Z' diff --git a/qai_hub_models/models/ffnet_40s/perf.yaml b/qai_hub_models/models/ffnet_40s/perf.yaml index d3007653..308f1f0b 100644 --- a/qai_hub_models/models/ffnet_40s/perf.yaml +++ b/qai_hub_models/models/ffnet_40s/perf.yaml @@ -45,11 +45,11 @@ models: - name: FFNet-40S performance_metrics: - torchscript_onnx_tflite: - inference_time: 16996.0 - throughput: 58.83737349964698 + inference_time: 17041.0 + throughput: 58.68200222991609 estimated_peak_memory_range: - min: 3907584 - max: 6044232 + min: 2170880 + max: 4550112 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: j7gj3y48p + job_id: j1p3r89zp job_status: Passed torchscript_onnx_qnn: - inference_time: 17404.0 - throughput: 57.45805561939784 + inference_time: 17425.0 + throughput: 57.38880918220947 estimated_peak_memory_range: - min: 25206784 - max: 51404216 + min: 25223168 + max: 41762760 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: jz57o8rrg + job_id: jmg9q8mvp job_status: Passed torchscript_onnx: - inference_time: 26886.0 - throughput: 37.19407870267054 + inference_time: 26795.0 + throughput: 37.32039559619332 estimated_peak_memory_range: - min: 25268224 - max: 493740712 + min: 25255936 + max: 27765408 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 142 - job_id: j2p0ox0ep + job_id: jopr79leg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:42:51Z' + timestamp: '2024-08-27T00:55:46Z' - torchscript_onnx_tflite: - inference_time: 12327.0 - throughput: 81.12273870365864 + inference_time: 12298.0 + throughput: 81.3140348024069 estimated_peak_memory_range: - min: 647168 - max: 94964064 + min: 2027520 + max: 95591424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: jlpe6x30g + job_id: jwgo9mrdg job_status: Passed torchscript_onnx_qnn: - inference_time: 12563.0 - throughput: 79.59882193743533 + inference_time: 12478.0 + throughput: 80.14104824491105 estimated_peak_memory_range: - min: 25198592 - max: 57765024 + min: 25202688 + max: 57333344 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: jqp4e2rlg + job_id: jnp1m3jlp job_status: Passed torchscript_onnx: - inference_time: 19488.0 - throughput: 51.3136288998358 + inference_time: 19377.0 + throughput: 51.607575992155645 estimated_peak_memory_range: - min: 31268864 - max: 141226960 + min: 31633408 + max: 141809936 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 142 - job_id: j1p8jky85 + job_id: jep2zjrmp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:42:52Z' + timestamp: '2024-08-27T00:55:47Z' - torchscript_onnx_tflite: - inference_time: 17000.0 - throughput: 58.8235294117647 + inference_time: 16935.0 + throughput: 59.04930617065249 estimated_peak_memory_range: - min: 2539520 - max: 374747584 + min: 2523136 + max: 34545512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: jygzzyk6g + job_id: j1pvn4dmg job_status: Passed torchscript_onnx_qnn: - inference_time: 17906.0 - throughput: 55.847202055177036 + inference_time: 17430.0 + throughput: 57.37234652897303 estimated_peak_memory_range: - min: 25251840 - max: 26631784 + min: 25247744 + max: 26673248 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: jo5m9lxqg + job_id: jz5786erp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:42:45Z' + timestamp: '2024-08-27T00:55:41Z' - torchscript_onnx_tflite: - inference_time: 27816.0 - throughput: 35.950532067874605 + inference_time: 27714.0 + throughput: 36.08284621490943 estimated_peak_memory_range: - min: 2580480 - max: 87222592 + min: 2727936 + max: 87559792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: jz5wyznjg + job_id: j7gj81785 job_status: Passed torchscript_onnx_qnn: - inference_time: 28148.0 - throughput: 35.526502771067214 + inference_time: 28449.0 + throughput: 35.15062040845021 estimated_peak_memory_range: - min: 25198592 - max: 53854880 + min: 25182208 + max: 54087584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: jqpy8y34g + job_id: jegnwxzmg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:42:50Z' + timestamp: '2024-08-27T00:55:45Z' - torchscript_onnx_tflite: - inference_time: 17080.0 - throughput: 58.5480093676815 + inference_time: 16982.0 + throughput: 58.885879166175954 estimated_peak_memory_range: - min: 2531328 - max: 4769912 + min: 2535424 + max: 4850984 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: jmg9o2evg + job_id: jlpen2z0p job_status: Passed torchscript_onnx_qnn: - inference_time: 17644.0 - throughput: 56.67649059170256 + inference_time: 17714.0 + throughput: 56.452523427797225 estimated_peak_memory_range: - min: 25247744 - max: 26551304 + min: 25260032 + max: 26642544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: jegn1wvmp + job_id: jqp428ylg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:42:46Z' + timestamp: '2024-08-27T00:55:42Z' - torchscript_onnx_tflite: - inference_time: 17018.0 - throughput: 58.76131155247385 + inference_time: 16946.0 + throughput: 59.01097604154373 estimated_peak_memory_range: - min: 122880 - max: 2772648 + min: 2572288 + max: 4543984 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: jnp1o1xl5 + job_id: jygz0wm65 job_status: Passed torchscript_onnx_qnn: - inference_time: 17887.0 - throughput: 55.906524291384805 + inference_time: 17705.0 + throughput: 56.481219994351875 estimated_peak_memory_range: - min: 25239552 - max: 28912160 + min: 25251840 + max: 26472088 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: joprx73ep + job_id: j0pxzml95 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:42:48Z' + timestamp: '2024-08-27T00:55:43Z' - torchscript_onnx_tflite: - inference_time: 17037.0 - throughput: 58.69577977343429 + inference_time: 16947.0 + throughput: 59.00749395173187 estimated_peak_memory_range: - min: 2584576 - max: 4728504 + min: 2551808 + max: 4508504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: jvgd64llp + job_id: jz5wrx7jp job_status: Passed torchscript_onnx_qnn: - inference_time: 17893.0 - throughput: 55.88777734309507 + inference_time: 17434.0 + throughput: 57.359183205231155 estimated_peak_memory_range: - min: 25243648 - max: 26723432 + min: 25223168 + max: 26859272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: jep2ozymg + job_id: jo5ml40qg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:42:49Z' + timestamp: '2024-08-27T00:55:44Z' - torchscript_onnx_qnn: - inference_time: 17800.0 - throughput: 56.17977528089887 + inference_time: 17849.0 + throughput: 56.025547649728274 estimated_peak_memory_range: min: 25219072 max: 25219072 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: j0px0zo9p + job_id: jvgdm03lg job_status: Passed torchscript_onnx: - inference_time: 26275.0 - throughput: 38.05899143672693 + inference_time: 26173.0 + throughput: 38.20731287968517 estimated_peak_memory_range: - min: 25223168 - max: 25223168 + min: 25219072 + max: 25219072 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 142 - job_id: jogk6kxo5 + job_id: jqpyyno4p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:42:53Z' + timestamp: '2024-08-27T00:55:48Z' diff --git a/qai_hub_models/models/ffnet_40s_quantized/perf.yaml b/qai_hub_models/models/ffnet_40s_quantized/perf.yaml index 3ca5f2a3..424f8072 100644 --- a/qai_hub_models/models/ffnet_40s_quantized/perf.yaml +++ b/qai_hub_models/models/ffnet_40s_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: FFNet-40S-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 4097.0 - throughput: 244.081034903588 + inference_time: 4101.0 + throughput: 243.84296513045598 estimated_peak_memory_range: - min: 647168 - max: 5324392 + min: 32768 + max: 8079096 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: j7gj3ym8p + job_id: j7gj81q75 job_status: Passed torchscript_onnx: - inference_time: 8896.0 - throughput: 112.41007194244604 + inference_time: 9022.0 + throughput: 110.84016847705608 estimated_peak_memory_range: - min: 7901184 - max: 221513128 + min: 6328320 + max: 15672208 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,7 +78,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: jn5q41omg + job_id: j2p0xkjep job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -87,13 +87,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:42:07Z' + timestamp: '2024-08-27T00:55:06Z' - torchscript_onnx_tflite: - inference_time: 2936.0 - throughput: 340.59945504087193 + inference_time: 2919.0 + throughput: 342.58307639602606 estimated_peak_memory_range: - min: 634880 - max: 62853872 + min: 643072 + max: 64439184 primary_compute_unit: NPU precision: int8 layer_info: @@ -101,14 +101,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jlpe6x10g + job_id: jygz0wnz5 job_status: Passed torchscript_onnx: - inference_time: 6301.0 - throughput: 158.70496746548167 + inference_time: 6282.0 + throughput: 159.1849729385546 estimated_peak_memory_range: - min: 7626752 - max: 91998800 + min: 7585792 + max: 93256960 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,7 +116,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: j1glw8rlp + job_id: j1p8k8x8p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -125,13 +125,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:42:08Z' + timestamp: '2024-08-27T00:55:07Z' - torchscript_onnx_tflite: - inference_time: 4137.0 - throughput: 241.72105390379502 + inference_time: 4036.0 + throughput: 247.7700693756194 estimated_peak_memory_range: - min: 651264 - max: 2143144 + min: 688128 + max: 171069568 primary_compute_unit: NPU precision: int8 layer_info: @@ -139,7 +139,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jygzzy96g + job_id: jz5wrx4zp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -148,13 +148,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:41:48Z' + timestamp: '2024-08-27T00:54:50Z' - torchscript_onnx_tflite: - inference_time: 5146.0 - throughput: 194.32568985619898 + inference_time: 5062.0 + throughput: 197.55037534571315 estimated_peak_memory_range: min: 638976 - max: 64269040 + max: 66085664 primary_compute_unit: NPU precision: int8 layer_info: @@ -162,7 +162,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jz5wyzvjg + job_id: jmg9q8dqp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -171,13 +171,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:41:49Z' + timestamp: '2024-08-27T00:54:51Z' - torchscript_onnx_tflite: - inference_time: 4098.0 - throughput: 244.0214738897023 + inference_time: 4143.0 + throughput: 241.3709872073377 estimated_peak_memory_range: - min: 651264 - max: 2195768 + min: 688128 + max: 189233040 primary_compute_unit: NPU precision: int8 layer_info: @@ -185,7 +185,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jmg9o21vg + job_id: jnp1m36kp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -194,13 +194,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:41:50Z' + timestamp: '2024-08-27T00:54:52Z' - torchscript_onnx_tflite: - inference_time: 4120.0 - throughput: 242.71844660194174 + inference_time: 4061.0 + throughput: 246.2447672986949 estimated_peak_memory_range: - min: 643072 - max: 15259024 + min: 659456 + max: 2181256 primary_compute_unit: NPU precision: int8 layer_info: @@ -208,7 +208,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jnp1o1ll5 + job_id: jvgdm02kg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -217,13 +217,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:41:51Z' + timestamp: '2024-08-27T00:54:53Z' - torchscript_onnx_tflite: - inference_time: 4102.0 - throughput: 243.78352023403218 + inference_time: 4169.0 + throughput: 239.86567522187576 estimated_peak_memory_range: - min: 32768 - max: 184360312 + min: 675840 + max: 5423576 primary_compute_unit: NPU precision: int8 layer_info: @@ -231,7 +231,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jvgd649lp + job_id: jz5wrx4jp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -240,13 +240,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:41:52Z' + timestamp: '2024-08-27T00:54:54Z' - torchscript_onnx_tflite: - inference_time: 26326.0 - throughput: 37.98526171845324 + inference_time: 26250.0 + throughput: 38.095238095238095 estimated_peak_memory_range: - min: 700416 - max: 39694256 + min: 729088 + max: 40850592 primary_compute_unit: NPU precision: int8 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jz57on3rg + job_id: jmg9q8dvp job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:41:54Z' + timestamp: '2024-08-27T00:54:55Z' - torchscript_onnx_tflite: - inference_time: 190540.0 - throughput: 5.24824183898394 + inference_time: 188112.0 + throughput: 5.315981968189164 estimated_peak_memory_range: - min: 819200 - max: 11045848 + min: 770048 + max: 7386752 primary_compute_unit: NPU precision: int8 layer_info: @@ -277,7 +277,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 99 - job_id: jqp4e40lg + job_id: jnp1m36lp job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -286,13 +286,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:41:55Z' + timestamp: '2024-08-27T00:54:56Z' - torchscript_onnx: - inference_time: 8506.0 - throughput: 117.56407241946862 + inference_time: 8431.0 + throughput: 118.60989206499822 estimated_peak_memory_range: - min: 10461184 - max: 10461184 + min: 10342400 + max: 10342400 primary_compute_unit: NPU precision: int8 layer_info: @@ -300,7 +300,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 92 - job_id: jw56oml75 + job_id: jogkkd4og job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -309,4 +309,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:42:09Z' + timestamp: '2024-08-27T00:55:08Z' diff --git a/qai_hub_models/models/ffnet_54s/perf.yaml b/qai_hub_models/models/ffnet_54s/perf.yaml index 7a0c9526..eebf6943 100644 --- a/qai_hub_models/models/ffnet_54s/perf.yaml +++ b/qai_hub_models/models/ffnet_54s/perf.yaml @@ -45,11 +45,11 @@ models: - name: FFNet-54S performance_metrics: - torchscript_onnx_tflite: - inference_time: 19912.0 - throughput: 50.220972278023304 + inference_time: 20119.0 + throughput: 49.70425965505244 estimated_peak_memory_range: - min: 2539520 - max: 47097024 + min: 663552 + max: 2488840 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jvgd64wkp + job_id: j1pvn497g job_status: Passed torchscript_onnx_qnn: - inference_time: 20147.0 - throughput: 49.635181416588075 + inference_time: 20099.0 + throughput: 49.75371909050202 estimated_peak_memory_range: - min: 25182208 - max: 45757648 + min: 23478272 + max: 45967488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 175 - job_id: j0px0rj9p + job_id: jvgdm0zkg job_status: Passed torchscript_onnx: - inference_time: 29746.0 - throughput: 33.617965440731524 + inference_time: 29784.0 + throughput: 33.5750738651625 estimated_peak_memory_range: - min: 25395200 - max: 28056264 + min: 25247744 + max: 40479016 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 177 - job_id: jogk60no5 + job_id: jqpyynmrp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:41:02Z' + timestamp: '2024-08-27T00:54:05Z' - torchscript_onnx_tflite: - inference_time: 14678.0 - throughput: 68.12917291184085 + inference_time: 14531.0 + throughput: 68.81838827334664 estimated_peak_memory_range: - min: 720896 - max: 103103728 + min: 2146304 + max: 103981424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jz5wyzojg + job_id: j7gj81w75 job_status: Passed torchscript_onnx_qnn: - inference_time: 14525.0 - throughput: 68.84681583476764 + inference_time: 14528.0 + throughput: 68.83259911894272 estimated_peak_memory_range: - min: 21004288 - max: 55578320 + min: 21069824 + max: 56785120 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 175 - job_id: jo5m9k2qg + job_id: jz57867qp job_status: Passed torchscript_onnx: - inference_time: 21875.0 - throughput: 45.714285714285715 + inference_time: 21867.0 + throughput: 45.73101019801528 estimated_peak_memory_range: - min: 33140736 - max: 153616112 + min: 2183168 + max: 123621904 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 177 - job_id: jn5q41kmg + job_id: j2p0xk62p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:41:03Z' + timestamp: '2024-08-27T00:54:06Z' - torchscript_onnx_tflite: - inference_time: 19991.0 - throughput: 50.0225101295583 + inference_time: 19690.0 + throughput: 50.787201625190455 estimated_peak_memory_range: - min: 2543616 - max: 5030784 + min: 2535424 + max: 5322488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jmg9o2vvg + job_id: jlpen2l7p job_status: Passed torchscript_onnx_qnn: - inference_time: 20044.0 - throughput: 49.89024146876871 + inference_time: 19915.0 + throughput: 50.21340697966357 estimated_peak_memory_range: - min: 25300992 - max: 29007944 + min: 25243648 + max: 26820208 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 175 - job_id: joprxdqep + job_id: j0pxzmdj5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:40:56Z' + timestamp: '2024-08-27T00:54:00Z' - torchscript_onnx_tflite: - inference_time: 32272.0 - throughput: 30.98661378284581 + inference_time: 32394.0 + throughput: 30.869914181638574 estimated_peak_memory_range: - min: 2572288 - max: 93265072 + min: 2527232 + max: 91220256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jnp1o10l5 + job_id: jygz0w4z5 job_status: Passed torchscript_onnx_qnn: - inference_time: 32971.0 - throughput: 30.32968366139941 + inference_time: 32928.0 + throughput: 30.369290573372204 estimated_peak_memory_range: min: 25235456 - max: 57321856 + max: 55396240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 175 - job_id: j1p8jr985 + job_id: jep2zjexp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:41:01Z' + timestamp: '2024-08-27T00:54:04Z' - torchscript_onnx_tflite: - inference_time: 19831.0 - throughput: 50.426100549644495 + inference_time: 19734.0 + throughput: 50.67396371744198 estimated_peak_memory_range: - min: 53248 - max: 2402416 + min: 2539520 + max: 4651064 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jvgd64wlp + job_id: jz5wrx1zp job_status: Passed torchscript_onnx_qnn: - inference_time: 20196.0 - throughput: 49.51475539710834 + inference_time: 20200.0 + throughput: 49.504950495049506 estimated_peak_memory_range: - min: 25251840 - max: 26581952 + min: 25280512 + max: 26732760 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 175 - job_id: jep2od6mg + job_id: jo5ml4dyg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:40:57Z' + timestamp: '2024-08-27T00:54:01Z' - torchscript_onnx_tflite: - inference_time: 19731.0 - throughput: 50.681668440525065 + inference_time: 20136.0 + throughput: 49.66229638458482 estimated_peak_memory_range: - min: 61440 - max: 2368216 + min: 2535424 + max: 4655280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jz57onwrg + job_id: jmg9q8xqp job_status: Passed torchscript_onnx_qnn: - inference_time: 20553.0 - throughput: 48.654697611054345 + inference_time: 20227.0 + throughput: 49.43886883868097 estimated_peak_memory_range: - min: 25264128 - max: 26535880 + min: 25260032 + max: 26482296 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 175 - job_id: jqpy82w4g + job_id: jegnwxkvg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:40:59Z' + timestamp: '2024-08-27T00:54:02Z' - torchscript_onnx_tflite: - inference_time: 19850.0 - throughput: 50.377833753148614 + inference_time: 20315.0 + throughput: 49.22471080482402 estimated_peak_memory_range: - min: 2551808 - max: 4972072 + min: 0 + max: 2988952 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jqp4e4olg + job_id: jnp1m3vkp job_status: Passed torchscript_onnx_qnn: - inference_time: 20108.0 - throughput: 49.73145016908693 + inference_time: 19971.0 + throughput: 50.0726052776526 estimated_peak_memory_range: - min: 25235456 - max: 29317832 + min: 25227264 + max: 26541544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 175 - job_id: j2p0o9qep + job_id: jopr79wvg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:41:00Z' + timestamp: '2024-08-27T00:54:03Z' - torchscript_onnx_qnn: - inference_time: 20310.0 - throughput: 49.23682914820286 + inference_time: 20355.0 + throughput: 49.12797838368951 estimated_peak_memory_range: min: 25219072 max: 25219072 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 175 - job_id: jegn1qymp + job_id: jqp4289qg job_status: Passed torchscript_onnx: - inference_time: 29734.0 - throughput: 33.63153292527073 + inference_time: 29394.0 + throughput: 34.02054841124039 estimated_peak_memory_range: - min: 25223168 - max: 25223168 + min: 25219072 + max: 25219072 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 177 - job_id: j1glw8zlp + job_id: j1p8k81zp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:41:05Z' + timestamp: '2024-08-27T00:54:07Z' diff --git a/qai_hub_models/models/ffnet_54s_quantized/perf.yaml b/qai_hub_models/models/ffnet_54s_quantized/perf.yaml index 15e38218..664b52aa 100644 --- a/qai_hub_models/models/ffnet_54s_quantized/perf.yaml +++ b/qai_hub_models/models/ffnet_54s_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: FFNet-54S-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 4739.0 - throughput: 211.01498206372653 + inference_time: 4770.0 + throughput: 209.64360587002096 estimated_peak_memory_range: - min: 651264 - max: 6022328 + min: 16384 + max: 22854432 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: jvgd64jkp + job_id: j1p3r8xxp job_status: Passed torchscript_onnx: - inference_time: 9440.0 - throughput: 105.9322033898305 + inference_time: 9549.0 + throughput: 104.72300764477956 estimated_peak_memory_range: - min: 36864 - max: 15976808 + min: 45056 + max: 15739816 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,7 +78,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: j7gj3yx7p + job_id: j1p8k84zp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -87,13 +87,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:40:17Z' + timestamp: '2024-08-27T00:53:23Z' - torchscript_onnx_tflite: - inference_time: 3378.0 - throughput: 296.0331557134399 + inference_time: 3365.0 + throughput: 297.1768202080238 estimated_peak_memory_range: - min: 643072 - max: 70670416 + min: 659456 + max: 72311472 primary_compute_unit: NPU precision: int8 layer_info: @@ -101,14 +101,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: jz57onqqg + job_id: jwgo9mo4g job_status: Passed torchscript_onnx: - inference_time: 6719.0 - throughput: 148.8316713796696 + inference_time: 6753.0 + throughput: 148.08233377758035 estimated_peak_memory_range: - min: 7626752 - max: 102122208 + min: 7569408 + max: 103346192 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,7 +116,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jlpe6x97g + job_id: jogkkd9yg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -125,13 +125,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:40:18Z' + timestamp: '2024-08-27T00:53:24Z' - torchscript_onnx_tflite: - inference_time: 4665.0 - throughput: 214.36227224008576 + inference_time: 4702.0 + throughput: 212.6754572522331 estimated_peak_memory_range: - min: 663552 - max: 23964776 + min: 643072 + max: 10691032 primary_compute_unit: NPU precision: int8 layer_info: @@ -139,7 +139,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: jqp4e4zqg + job_id: j1pvn4e7g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -148,13 +148,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:39:58Z' + timestamp: '2024-08-27T00:53:07Z' - torchscript_onnx_tflite: - inference_time: 5885.0 - throughput: 169.92353440951572 + inference_time: 5778.0 + throughput: 173.07026652821045 estimated_peak_memory_range: - min: 675840 - max: 72368768 + min: 647168 + max: 74739424 primary_compute_unit: NPU precision: int8 layer_info: @@ -162,7 +162,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: j0px0rwjp + job_id: j7gj81o75 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -171,13 +171,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:39:59Z' + timestamp: '2024-08-27T00:53:08Z' - torchscript_onnx_tflite: - inference_time: 4737.0 - throughput: 211.10407430863415 + inference_time: 4708.0 + throughput: 212.40441801189465 estimated_peak_memory_range: min: 24576 - max: 203474112 + max: 203938808 primary_compute_unit: NPU precision: int8 layer_info: @@ -185,7 +185,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: jo5m9kjyg + job_id: jlpen287p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -194,13 +194,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:40:00Z' + timestamp: '2024-08-27T00:53:09Z' - torchscript_onnx_tflite: - inference_time: 4733.0 - throughput: 211.28248468201986 + inference_time: 4789.0 + throughput: 208.81186051367717 estimated_peak_memory_range: - min: 36864 - max: 202783304 + min: 659456 + max: 13000672 primary_compute_unit: NPU precision: int8 layer_info: @@ -208,7 +208,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: jegn1q2vp + job_id: jygz0w8z5 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -217,13 +217,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:40:01Z' + timestamp: '2024-08-27T00:53:10Z' - torchscript_onnx_tflite: - inference_time: 4864.0 - throughput: 205.5921052631579 + inference_time: 4711.0 + throughput: 212.26915729144557 estimated_peak_memory_range: - min: 45056 - max: 204024560 + min: 548864 + max: 5569712 primary_compute_unit: NPU precision: int8 layer_info: @@ -231,7 +231,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: joprxdkvp + job_id: jz5wrx8zp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -240,13 +240,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:40:02Z' + timestamp: '2024-08-27T00:53:11Z' - torchscript_onnx_tflite: - inference_time: 30210.0 - throughput: 33.101621979476995 + inference_time: 29997.0 + throughput: 33.33666700003334 estimated_peak_memory_range: - min: 684032 - max: 42435488 + min: 708608 + max: 45621840 primary_compute_unit: NPU precision: int8 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: jep2od8xg + job_id: jmg9q8kqp job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:40:03Z' + timestamp: '2024-08-27T00:53:12Z' - torchscript_onnx_tflite: - inference_time: 202690.0 - throughput: 4.933642508263851 + inference_time: 201260.0 + throughput: 4.968697207592169 estimated_peak_memory_range: - min: 999424 - max: 2987960 + min: 823296 + max: 11537056 primary_compute_unit: NPU precision: int8 layer_info: @@ -277,7 +277,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 120 - job_id: jqpy82erg + job_id: jnp1m37kp job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -286,13 +286,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:40:04Z' + timestamp: '2024-08-27T00:53:13Z' - torchscript_onnx: - inference_time: 9068.0 - throughput: 110.27790030877811 + inference_time: 9043.0 + throughput: 110.58277120424638 estimated_peak_memory_range: - min: 14503936 - max: 14503936 + min: 15859712 + max: 15859712 primary_compute_unit: NPU precision: int8 layer_info: @@ -300,7 +300,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jygzzyezg + job_id: jn5qdwm7g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -309,4 +309,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:40:19Z' + timestamp: '2024-08-27T00:53:25Z' diff --git a/qai_hub_models/models/ffnet_78s/perf.yaml b/qai_hub_models/models/ffnet_78s/perf.yaml index 63ad5def..e1e07743 100644 --- a/qai_hub_models/models/ffnet_78s/perf.yaml +++ b/qai_hub_models/models/ffnet_78s/perf.yaml @@ -45,11 +45,11 @@ models: - name: FFNet-78S performance_metrics: - torchscript_onnx_tflite: - inference_time: 23466.0 - throughput: 42.61484701269922 + inference_time: 25922.0 + throughput: 38.57727027235553 estimated_peak_memory_range: - min: 2134016 - max: 4201936 + min: 2576384 + max: 5186496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jqp4e47qg + job_id: j1pvn447g job_status: Passed torchscript_onnx_qnn: - inference_time: 24452.0 - throughput: 40.89645018812367 + inference_time: 23887.0 + throughput: 41.86377527525432 estimated_peak_memory_range: - min: 25526272 - max: 46029800 + min: 29065216 + max: 49463536 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: j2p0o9n2p + job_id: jvgdm00kg job_status: Passed torchscript_onnx: - inference_time: 33750.0 - throughput: 29.62962962962963 + inference_time: 34248.0 + throughput: 29.19878533053025 estimated_peak_memory_range: - min: 25247744 - max: 495051688 + min: 26062848 + max: 57577088 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 237 - job_id: j1pv2mj7g + job_id: jqpyynqrp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:39:06Z' + timestamp: '2024-08-27T00:52:20Z' - torchscript_onnx_tflite: - inference_time: 17351.0 - throughput: 57.633565788715345 + inference_time: 17284.0 + throughput: 57.85697755149271 estimated_peak_memory_range: - min: 12288 - max: 114999760 + min: 1818624 + max: 117883632 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: j0px0rqjp + job_id: j7gj81175 job_status: Passed torchscript_onnx_qnn: - inference_time: 17820.0 - throughput: 56.11672278338945 + inference_time: 17579.0 + throughput: 56.88605722737357 estimated_peak_memory_range: - min: 23179264 - max: 60667968 + min: 19968000 + max: 56127360 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: j1p8jrlz5 + job_id: jz57866qp job_status: Passed torchscript_onnx: - inference_time: 24951.0 - throughput: 40.07855396577291 + inference_time: 24928.0 + throughput: 40.11553273427471 estimated_peak_memory_range: - min: 31952896 - max: 168119760 + min: 2449408 + max: 139721232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 237 - job_id: j7gj3yj7p + job_id: j2p0xkd2p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:39:08Z' + timestamp: '2024-08-27T00:52:21Z' - torchscript_onnx_tflite: - inference_time: 23476.0 - throughput: 42.59669449650707 + inference_time: 23240.0 + throughput: 43.029259896729776 estimated_peak_memory_range: - min: 2568192 - max: 5362168 + min: 2441216 + max: 4729928 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jo5m9k7yg + job_id: jlpen227p job_status: Passed torchscript_onnx_qnn: - inference_time: 23658.0 - throughput: 42.268999915462 + inference_time: 23484.0 + throughput: 42.582183614375744 estimated_peak_memory_range: - min: 25247744 - max: 26427736 + min: 25292800 + max: 27040624 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jn5q41j7g + job_id: j0pxzmmj5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:39:00Z' + timestamp: '2024-08-27T00:52:15Z' - torchscript_onnx_tflite: - inference_time: 42216.0 - throughput: 23.687701345461438 + inference_time: 38863.0 + throughput: 25.73141548516584 estimated_peak_memory_range: - min: 2596864 - max: 102467760 + min: 2584576 + max: 101496192 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jegn1qjvp + job_id: jygz0wwz5 job_status: Passed torchscript_onnx_qnn: - inference_time: 39626.0 - throughput: 25.235956190380055 + inference_time: 39293.0 + throughput: 25.44982566869417 estimated_peak_memory_range: - min: 25210880 - max: 58785984 + min: 25231360 + max: 58810464 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jwgodwj45 + job_id: jep2zj4xp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:39:05Z' + timestamp: '2024-08-27T00:52:19Z' - torchscript_onnx_tflite: - inference_time: 23327.0 - throughput: 42.868778668495736 + inference_time: 23308.0 + throughput: 42.90372404324695 estimated_peak_memory_range: min: 2535424 - max: 4748312 + max: 4533128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: joprxdzvp + job_id: jz5wrxxzp job_status: Passed torchscript_onnx_qnn: - inference_time: 23906.0 - throughput: 41.830502802643686 + inference_time: 24061.0 + throughput: 41.561032376044224 estimated_peak_memory_range: - min: 25284608 - max: 26732680 + min: 25321472 + max: 26597632 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: j1glw8jep + job_id: jo5ml44yg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:39:01Z' + timestamp: '2024-08-27T00:52:16Z' - torchscript_onnx_tflite: - inference_time: 23246.0 - throughput: 43.018153660844874 + inference_time: 23163.0 + throughput: 43.17230065190174 estimated_peak_memory_range: - min: 2564096 - max: 4966672 + min: 77824 + max: 692830792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jep2od2xg + job_id: jmg9q88qp job_status: Passed torchscript_onnx_qnn: - inference_time: 24132.0 - throughput: 41.43875352229405 + inference_time: 23863.0 + throughput: 41.9058793948791 estimated_peak_memory_range: - min: 25280512 - max: 26583512 + min: 25313280 + max: 26855648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jw56omkv5 + job_id: jegnwxovg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:39:03Z' + timestamp: '2024-08-27T00:52:17Z' - torchscript_onnx_tflite: - inference_time: 23422.0 - throughput: 42.6949022286739 + inference_time: 23268.0 + throughput: 42.9774798005845 estimated_peak_memory_range: - min: 2535424 - max: 5101592 + min: 2547712 + max: 4737280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jqpy829rg + job_id: jnp1m33kp job_status: Passed torchscript_onnx_qnn: - inference_time: 23571.0 - throughput: 42.42501378812948 + inference_time: 23894.0 + throughput: 41.851510839541305 estimated_peak_memory_range: - min: 25288704 - max: 26933488 + min: 25280512 + max: 26796936 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: j1p3o7yxp + job_id: jopr79ovg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:39:04Z' + timestamp: '2024-08-27T00:52:18Z' - torchscript_onnx_qnn: - inference_time: 24237.0 - throughput: 41.25923175310476 + inference_time: 24201.0 + throughput: 41.32060658650469 estimated_peak_memory_range: min: 25219072 max: 25219072 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 235 - job_id: jogk60jy5 + job_id: jqp4288qg job_status: Passed torchscript_onnx: - inference_time: 33290.0 - throughput: 30.039050765995796 + inference_time: 33668.0 + throughput: 29.701793988356897 estimated_peak_memory_range: - min: 34783232 - max: 34783232 + min: 34734080 + max: 34734080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 237 - job_id: jlpe6xj7g + job_id: j1p8k86zp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:39:09Z' + timestamp: '2024-08-27T00:52:22Z' diff --git a/qai_hub_models/models/ffnet_78s_lowres/perf.yaml b/qai_hub_models/models/ffnet_78s_lowres/perf.yaml index fab82940..8e733658 100644 --- a/qai_hub_models/models/ffnet_78s_lowres/perf.yaml +++ b/qai_hub_models/models/ffnet_78s_lowres/perf.yaml @@ -45,11 +45,11 @@ models: - name: FFNet-78S-LowRes performance_metrics: - torchscript_onnx_tflite: - inference_time: 7336.0 - throughput: 136.31406761177755 + inference_time: 7372.0 + throughput: 135.6483993488877 estimated_peak_memory_range: - min: 651264 - max: 2676000 + min: 667648 + max: 2870400 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jo5m9k1yg + job_id: jmg9q89mp job_status: Passed torchscript_onnx_qnn: - inference_time: 7659.0 - throughput: 130.56534795665232 + inference_time: 7849.0 + throughput: 127.40476493820869 estimated_peak_memory_range: - min: 6307840 - max: 26020200 + min: 6311936 + max: 17652256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 236 - job_id: jogk601y5 + job_id: jz5786vqp job_status: Passed torchscript_onnx: - inference_time: 8817.0 - throughput: 113.41726210729273 + inference_time: 8814.0 + throughput: 113.45586566825504 estimated_peak_memory_range: - min: 24576 - max: 53859248 + min: 49152 + max: 53788712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 238 - job_id: jlpe6xq7g + job_id: j1p8k88zp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:38:19Z' + timestamp: '2024-08-27T00:51:37Z' - torchscript_onnx_tflite: - inference_time: 5197.0 - throughput: 192.41870309794112 + inference_time: 5131.0 + throughput: 194.89378288832586 estimated_peak_memory_range: - min: 323584 - max: 60110480 + min: 638976 + max: 61707616 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jegn1q4vp + job_id: jnp1m3qnp job_status: Passed torchscript_onnx_qnn: - inference_time: 5472.0 - throughput: 182.7485380116959 + inference_time: 5403.0 + throughput: 185.08236165093467 estimated_peak_memory_range: - min: 6307840 - max: 30713648 + min: 6336512 + max: 29902768 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 236 - job_id: jn5q41n7g + job_id: jqp428jqg job_status: Passed torchscript_onnx: - inference_time: 6393.0 - throughput: 156.42108556233381 + inference_time: 6445.0 + throughput: 155.1590380139643 estimated_peak_memory_range: - min: 7622656 - max: 77756768 + min: 495616 + max: 71822800 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 238 - job_id: jygzzy6zg + job_id: jogkkddyg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:38:20Z' + timestamp: '2024-08-27T00:51:38Z' - torchscript_onnx_tflite: - inference_time: 7317.0 - throughput: 136.66803334700015 + inference_time: 7269.0 + throughput: 137.57050488375293 estimated_peak_memory_range: - min: 811008 - max: 2364096 + min: 663552 + max: 2774424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: joprxdrvp + job_id: jvgdm076g job_status: Passed torchscript_onnx_qnn: - inference_time: 7031.0 - throughput: 142.22727919214904 + inference_time: 7095.0 + throughput: 140.94432699083862 estimated_peak_memory_range: - min: 6328320 - max: 8047944 + min: 6352896 + max: 7715824 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 236 - job_id: jw56omxv5 + job_id: jo5ml4vyg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:38:13Z' + timestamp: '2024-08-27T00:51:32Z' - torchscript_onnx_tflite: - inference_time: 11212.0 - throughput: 89.19015340706386 + inference_time: 11140.0 + throughput: 89.76660682226212 estimated_peak_memory_range: - min: 3305472 - max: 57235472 + min: 655360 + max: 53357056 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jep2od1xg + job_id: jz5wrxmzp job_status: Passed torchscript_onnx_qnn: - inference_time: 11570.0 - throughput: 86.4304235090752 + inference_time: 11557.0 + throughput: 86.5276455827637 estimated_peak_memory_range: - min: 6307840 - max: 26776464 + min: 6242304 + max: 25461168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 236 - job_id: j7gj3y97p + job_id: j2p0xkk2p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:38:18Z' + timestamp: '2024-08-27T00:51:36Z' - torchscript_onnx_tflite: - inference_time: 7420.0 - throughput: 134.77088948787062 + inference_time: 7371.0 + throughput: 135.666802333469 estimated_peak_memory_range: - min: 675840 - max: 2473344 + min: 0 + max: 598112688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jqpy82lrg + job_id: jmg9q89qp job_status: Passed torchscript_onnx_qnn: - inference_time: 7041.0 - throughput: 142.02528049992898 + inference_time: 7064.0 + throughput: 141.56285390713478 estimated_peak_memory_range: - min: 6377472 - max: 7640912 + min: 6369280 + max: 7577448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 236 - job_id: j1p3o7dxp + job_id: jegnwxxvg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:38:14Z' + timestamp: '2024-08-27T00:51:33Z' - torchscript_onnx_tflite: - inference_time: 7448.0 - throughput: 134.2642320085929 + inference_time: 7244.0 + throughput: 138.04527885146328 estimated_peak_memory_range: - min: 638976 - max: 2420288 + min: 0 + max: 566022768 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: j2p0o9w2p + job_id: jnp1m3qkp job_status: Passed torchscript_onnx_qnn: - inference_time: 7148.0 - throughput: 139.89927252378288 + inference_time: 7188.0 + throughput: 139.1207568169171 estimated_peak_memory_range: - min: 6340608 - max: 8175280 + min: 6356992 + max: 7905312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 236 - job_id: jwgodwx45 + job_id: jopr799vg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:38:15Z' + timestamp: '2024-08-27T00:51:34Z' - torchscript_onnx_tflite: - inference_time: 7311.0 - throughput: 136.7801942278758 + inference_time: 7352.0 + throughput: 136.01741022850925 estimated_peak_memory_range: - min: 679936 - max: 2623624 + min: 696320 + max: 2511152 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: j1p8jrnz5 + job_id: jvgdm07kg job_status: Passed torchscript_onnx_qnn: - inference_time: 7115.0 - throughput: 140.54813773717498 + inference_time: 7139.0 + throughput: 140.07564084605687 estimated_peak_memory_range: - min: 6344704 - max: 7609920 + min: 6377472 + max: 7653384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 236 - job_id: j1pv2m87g + job_id: jep2zjjxp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:38:17Z' + timestamp: '2024-08-27T00:51:35Z' - torchscript_onnx_qnn: - inference_time: 7530.0 - throughput: 132.80212483399734 + inference_time: 7526.0 + throughput: 132.87270794578794 estimated_peak_memory_range: min: 6303744 max: 6303744 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 236 - job_id: j1glw8dep + job_id: j0pxzmej5 job_status: Passed torchscript_onnx: - inference_time: 8779.0 - throughput: 113.90818999886092 + inference_time: 8773.0 + throughput: 113.98609369656901 estimated_peak_memory_range: - min: 50298880 - max: 50298880 + min: 51597312 + max: 51597312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 238 - job_id: jz5wyzkzg + job_id: jn5qdww7g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:38:21Z' + timestamp: '2024-08-27T00:51:39Z' diff --git a/qai_hub_models/models/ffnet_78s_quantized/perf.yaml b/qai_hub_models/models/ffnet_78s_quantized/perf.yaml index f8de54bf..450c3abd 100644 --- a/qai_hub_models/models/ffnet_78s_quantized/perf.yaml +++ b/qai_hub_models/models/ffnet_78s_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: FFNet-78S-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 5764.0 - throughput: 173.49063150589868 + inference_time: 5790.0 + throughput: 172.71157167530225 estimated_peak_memory_range: - min: 643072 - max: 7790168 + min: 28672 + max: 23254456 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: jo5m9klyg + job_id: jmg9q83mp job_status: Passed torchscript_onnx: - inference_time: 9749.0 - throughput: 102.57462303826034 + inference_time: 9706.0 + throughput: 103.0290541932825 estimated_peak_memory_range: - min: 49152 - max: 23931976 + min: 28672 + max: 24692424 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,7 +78,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jmg9o27qg + job_id: jwgo9me1g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -87,13 +87,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:37:34Z' + timestamp: '2024-08-27T00:50:56Z' - torchscript_onnx_tflite: - inference_time: 4083.0 - throughput: 244.91795248591723 + inference_time: 4105.0 + throughput: 243.605359317905 estimated_peak_memory_range: min: 638976 - max: 84508496 + max: 86038416 primary_compute_unit: NPU precision: int8 layer_info: @@ -101,14 +101,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: jegn1qdvp + job_id: jnp1m3dnp job_status: Passed torchscript_onnx: - inference_time: 6792.0 - throughput: 147.23203769140164 + inference_time: 6899.0 + throughput: 144.94854326714017 estimated_peak_memory_range: - min: 7561216 - max: 120696288 + min: 7585792 + max: 123239776 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,7 +116,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jnp1o1kk5 + job_id: j1pvn4zzg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -125,13 +125,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:37:35Z' + timestamp: '2024-08-27T00:50:57Z' - torchscript_onnx_tflite: - inference_time: 5846.0 - throughput: 171.05713308244952 + inference_time: 5693.0 + throughput: 175.65431231336729 estimated_peak_memory_range: - min: 12288 - max: 278188160 + min: 659456 + max: 2346368 primary_compute_unit: NPU precision: int8 layer_info: @@ -139,7 +139,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: joprxdmvp + job_id: jvgdm0r6g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -148,13 +148,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:37:15Z' + timestamp: '2024-08-27T00:50:40Z' - torchscript_onnx_tflite: - inference_time: 7095.0 - throughput: 140.94432699083862 + inference_time: 7113.0 + throughput: 140.58765640376774 estimated_peak_memory_range: - min: 638976 - max: 87657392 + min: 655360 + max: 90685952 primary_compute_unit: NPU precision: int8 layer_info: @@ -162,7 +162,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: jep2odqxg + job_id: jz5786jnp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -171,13 +171,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:37:16Z' + timestamp: '2024-08-27T00:50:41Z' - torchscript_onnx_tflite: - inference_time: 5716.0 - throughput: 174.9475157452764 + inference_time: 5692.0 + throughput: 175.68517217146874 estimated_peak_memory_range: min: 647168 - max: 15083760 + max: 16649600 primary_compute_unit: NPU precision: int8 layer_info: @@ -185,7 +185,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: jqpy82krg + job_id: jqp428x2g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -194,13 +194,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:37:18Z' + timestamp: '2024-08-27T00:50:42Z' - torchscript_onnx_tflite: - inference_time: 5735.0 - throughput: 174.36791630340016 + inference_time: 5765.0 + throughput: 173.46053772766695 estimated_peak_memory_range: - min: 684032 - max: 6377200 + min: 647168 + max: 16173880 primary_compute_unit: NPU precision: int8 layer_info: @@ -208,7 +208,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: j2p0o982p + job_id: j0pxzm785 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -217,13 +217,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:37:19Z' + timestamp: '2024-08-27T00:50:43Z' - torchscript_onnx_tflite: - inference_time: 5720.0 - throughput: 174.82517482517483 + inference_time: 5692.0 + throughput: 175.68517217146874 estimated_peak_memory_range: - min: 704512 - max: 17432384 + min: 663552 + max: 17645504 primary_compute_unit: NPU precision: int8 layer_info: @@ -231,7 +231,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: j1p8jrdz5 + job_id: jo5ml4w7g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -240,13 +240,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:37:20Z' + timestamp: '2024-08-27T00:50:44Z' - torchscript_onnx_tflite: - inference_time: 35673.0 - throughput: 28.032405460712585 + inference_time: 35594.0 + throughput: 28.094622689217285 estimated_peak_memory_range: min: 12288 - max: 45579776 + max: 47541024 primary_compute_unit: NPU precision: int8 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: jogk60wy5 + job_id: jegnwx9jg job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:37:21Z' + timestamp: '2024-08-27T00:50:45Z' - torchscript_onnx_tflite: - inference_time: 220093.0 - throughput: 4.543533869773232 + inference_time: 219412.0 + throughput: 4.55763586312508 estimated_peak_memory_range: - min: 946176 - max: 3833064 + min: 929792 + max: 7997504 primary_compute_unit: NPU precision: int8 layer_info: @@ -277,7 +277,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 156 - job_id: jn5q41x7g + job_id: jopr794kg job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -286,13 +286,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:37:22Z' + timestamp: '2024-08-27T00:50:46Z' - torchscript_onnx: - inference_time: 9398.0 - throughput: 106.40561821664184 + inference_time: 9407.0 + throughput: 106.30381630700542 estimated_peak_memory_range: - min: 23748608 - max: 23748608 + min: 23801856 + max: 23801856 primary_compute_unit: NPU precision: int8 layer_info: @@ -300,7 +300,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jvgd64ykp + job_id: j7gj81k15 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -309,4 +309,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:37:37Z' + timestamp: '2024-08-27T00:50:58Z' diff --git a/qai_hub_models/models/googlenet/perf.yaml b/qai_hub_models/models/googlenet/perf.yaml index a117a400..988f2a23 100644 --- a/qai_hub_models/models/googlenet/perf.yaml +++ b/qai_hub_models/models/googlenet/perf.yaml @@ -45,11 +45,11 @@ models: - name: GoogLeNet performance_metrics: - torchscript_onnx_tflite: - inference_time: 1048.0 - throughput: 954.1984732824427 + inference_time: 1021.0 + throughput: 979.4319294809011 estimated_peak_memory_range: - min: 12288 - max: 2096792 + min: 28672 + max: 192513688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 84 - job_id: j1p8jrkq5 + job_id: jvgdm0v6g job_status: Passed torchscript_onnx_qnn: - inference_time: 1088.0 - throughput: 919.1176470588235 + inference_time: 1086.0 + throughput: 920.8103130755064 estimated_peak_memory_range: min: 16384 - max: 36357760 + max: 25143552 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 143 - job_id: j1pv2mnzg + job_id: jep2zjm6p job_status: Passed torchscript_onnx: - inference_time: 1280.0 - throughput: 781.25 + inference_time: 1251.0 + throughput: 799.3605115907275 estimated_peak_memory_range: - min: 28672 - max: 15663176 + min: 12288 + max: 15599800 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 145 - job_id: jz5wyzrzg + job_id: j1p3r84mp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:36:10Z' + timestamp: '2024-08-27T00:49:36Z' - torchscript_onnx_tflite: - inference_time: 644.0 - throughput: 1552.7950310559006 + inference_time: 633.0 + throughput: 1579.778830963665 estimated_peak_memory_range: - min: 16384 - max: 51086240 + min: 20480 + max: 51313328 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 84 - job_id: jogk60kv5 + job_id: jz5786dnp job_status: Passed torchscript_onnx_qnn: - inference_time: 686.0 - throughput: 1457.725947521866 + inference_time: 685.0 + throughput: 1459.85401459854 estimated_peak_memory_range: min: 618496 - max: 15868096 + max: 14109408 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 143 - job_id: j7gj3y81p + job_id: jqpyynd0p job_status: Passed torchscript_onnx: - inference_time: 830.0 - throughput: 1204.8192771084337 + inference_time: 823.0 + throughput: 1215.0668286755772 estimated_peak_memory_range: min: 0 - max: 52752000 + max: 55293712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 145 - job_id: jmg9o2qqg + job_id: jwgo9m11g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:36:11Z' + timestamp: '2024-08-27T00:49:37Z' - torchscript_onnx_tflite: - inference_time: 1040.0 - throughput: 961.5384615384615 + inference_time: 1016.0 + throughput: 984.2519685039371 estimated_peak_memory_range: - min: 16384 - max: 1233480 + min: 12288 + max: 41223608 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 84 - job_id: jn5q41deg + job_id: jqp428w2g job_status: Passed torchscript_onnx_qnn: - inference_time: 1066.0 - throughput: 938.0863039399625 + inference_time: 1065.0 + throughput: 938.9671361502348 estimated_peak_memory_range: min: 634880 - max: 2153856 + max: 2028368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 143 - job_id: jygzzy04g + job_id: j1p8k83qp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:36:04Z' + timestamp: '2024-08-27T00:49:31Z' - torchscript_onnx_tflite: - inference_time: 1536.0 - throughput: 651.0416666666666 + inference_time: 1479.0 + throughput: 676.132521974307 estimated_peak_memory_range: - min: 20480 - max: 51508832 + min: 16384 + max: 52198736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 84 - job_id: j1glw8q2p + job_id: j0pxzm185 job_status: Passed torchscript_onnx_qnn: - inference_time: 1583.0 - throughput: 631.7119393556538 + inference_time: 1558.0 + throughput: 641.8485237483953 estimated_peak_memory_range: min: 618496 - max: 19607680 + max: 19203168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 143 - job_id: jvgd64m6p + job_id: jw560v3n5 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:36:09Z' + timestamp: '2024-08-27T00:49:35Z' - torchscript_onnx_tflite: - inference_time: 1041.0 - throughput: 960.6147934678194 + inference_time: 1022.0 + throughput: 978.4735812133073 estimated_peak_memory_range: - min: 12288 - max: 18659608 + min: 24576 + max: 1360712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 84 - job_id: jw56om0n5 + job_id: jo5ml4z7g job_status: Passed torchscript_onnx_qnn: - inference_time: 1081.0 - throughput: 925.0693802035153 + inference_time: 1073.0 + throughput: 931.9664492078285 estimated_peak_memory_range: - min: 630784 - max: 1923464 + min: 638976 + max: 1854352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 143 - job_id: jz5wyzr4g + job_id: jogkkdlvg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:36:05Z' + timestamp: '2024-08-27T00:49:32Z' - torchscript_onnx_tflite: - inference_time: 1050.0 - throughput: 952.3809523809524 + inference_time: 1020.0 + throughput: 980.3921568627451 estimated_peak_memory_range: - min: 20480 - max: 1572968 + min: 24576 + max: 8820424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 84 - job_id: j1p3o7rmp + job_id: jegnwxejg job_status: Passed torchscript_onnx_qnn: - inference_time: 1076.0 - throughput: 929.368029739777 + inference_time: 1074.0 + throughput: 931.0986964618249 estimated_peak_memory_range: - min: 626688 - max: 1902720 + min: 643072 + max: 1846064 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 143 - job_id: jmg9o2qmg + job_id: jn5qdw7eg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:36:06Z' + timestamp: '2024-08-27T00:49:33Z' - torchscript_onnx_tflite: - inference_time: 1050.0 - throughput: 952.3809523809524 + inference_time: 1017.0 + throughput: 983.284169124877 estimated_peak_memory_range: - min: 12288 - max: 4517912 + min: 40960 + max: 193695056 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 84 - job_id: jwgodw915 + job_id: jopr79ykg job_status: Passed torchscript_onnx_qnn: - inference_time: 1074.0 - throughput: 931.0986964618249 + inference_time: 1064.0 + throughput: 939.8496240601504 estimated_peak_memory_range: - min: 626688 - max: 2045648 + min: 634880 + max: 1967976 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 143 - job_id: jnp1o1mn5 + job_id: j1glq702p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:36:07Z' + timestamp: '2024-08-27T00:49:34Z' - torchscript_onnx_qnn: - inference_time: 1230.0 - throughput: 813.0081300813008 + inference_time: 1243.0 + throughput: 804.5052292839904 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 143 - job_id: jlpe6xn8g + job_id: j2p0xk10p job_status: Passed torchscript_onnx: - inference_time: 1313.0 - throughput: 761.6146230007616 + inference_time: 1359.0 + throughput: 735.8351729212657 estimated_peak_memory_range: - min: 15601664 - max: 15601664 + min: 16752640 + max: 16752640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 145 - job_id: jnp1o1mk5 + job_id: j1pvn41zg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:36:12Z' + timestamp: '2024-08-27T00:49:38Z' diff --git a/qai_hub_models/models/googlenet_quantized/perf.yaml b/qai_hub_models/models/googlenet_quantized/perf.yaml index e505fa0b..95ce8c89 100644 --- a/qai_hub_models/models/googlenet_quantized/perf.yaml +++ b/qai_hub_models/models/googlenet_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: GoogLeNetQuantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 290.0 - throughput: 3448.2758620689656 + inference_time: 283.0 + throughput: 3533.5689045936397 estimated_peak_memory_range: min: 12288 - max: 1632736 + max: 97346344 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j2p0o990p + job_id: jz5wrxq4p job_status: Passed torchscript_onnx_qnn: - inference_time: 345.0 - throughput: 2898.550724637681 + inference_time: 332.0 + throughput: 3012.0481927710844 estimated_peak_memory_range: - min: 184320 - max: 9715120 + min: 32768 + max: 10151944 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,22 +78,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j7gj3yy1p - job_status: Passed - torchscript_onnx: - inference_time: 521.0 - throughput: 1919.3857965451057 - estimated_peak_memory_range: - min: 12288 - max: 9972976 - primary_compute_unit: NPU - precision: int8 - layer_info: - layers_on_npu: 91 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 91 - job_id: j0px0rr8p + job_id: jep2zjl6p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +87,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:35:30Z' + timestamp: '2024-08-27T00:48:53Z' - torchscript_onnx_tflite: - inference_time: 209.0 - throughput: 4784.688995215311 + inference_time: 197.0 + throughput: 5076.1421319796955 estimated_peak_memory_range: min: 12288 - max: 37666880 + max: 38146784 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +101,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j1p8jrrq5 + job_id: jnp1m3enp job_status: Passed torchscript_onnx_qnn: inference_time: 248.0 throughput: 4032.2580645161293 estimated_peak_memory_range: - min: 3416064 - max: 16534816 + min: 0 + max: 15199360 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jlpe6xx8g + job_id: jqpyyn60p job_status: Passed torchscript_onnx: - inference_time: 372.0 - throughput: 2688.1720430107525 + inference_time: 398.0 + throughput: 2512.5628140703516 estimated_peak_memory_range: - min: 12288 - max: 56079920 + min: 28672 + max: 57555312 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +131,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 91 - job_id: jo5m9kk7g + job_id: j1pvn4wzg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +140,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:35:31Z' + timestamp: '2024-08-27T00:49:02Z' - torchscript_onnx_tflite: - inference_time: 291.0 - throughput: 3436.426116838488 + inference_time: 272.0 + throughput: 3676.470588235294 estimated_peak_memory_range: min: 12288 - max: 1540920 + max: 3679672 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +154,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jogk600v5 + job_id: jvgdm0o6g job_status: Passed torchscript_onnx_qnn: - inference_time: 338.0 - throughput: 2958.579881656805 + inference_time: 339.0 + throughput: 2949.8525073746314 estimated_peak_memory_range: min: 176128 - max: 1432536 + max: 1684480 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +169,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jz5wyzz4g + job_id: j1p8k87qp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +178,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:35:23Z' + timestamp: '2024-08-27T00:48:56Z' - torchscript_onnx_tflite: - inference_time: 360.0 - throughput: 2777.777777777778 + inference_time: 337.0 + throughput: 2967.359050445104 estimated_peak_memory_range: min: 20480 - max: 37797552 + max: 39007824 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +192,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jn5q411eg + job_id: jz5786xnp job_status: Passed torchscript_onnx_qnn: - inference_time: 418.0 - throughput: 2392.3444976076553 + inference_time: 406.0 + throughput: 2463.054187192118 estimated_peak_memory_range: min: 159744 - max: 15673872 + max: 15407936 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +207,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jz57onnng + job_id: jw560v1n5 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +216,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:35:28Z' + timestamp: '2024-08-27T00:48:59Z' - torchscript_onnx_tflite: - inference_time: 289.0 - throughput: 3460.2076124567475 + inference_time: 280.0 + throughput: 3571.4285714285716 estimated_peak_memory_range: - min: 20480 - max: 1407000 + min: 16384 + max: 2219344 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +230,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j1glw882p + job_id: jqp428v2g job_status: Passed torchscript_onnx_qnn: - inference_time: 341.0 - throughput: 2932.551319648094 + inference_time: 339.0 + throughput: 2949.8525073746314 estimated_peak_memory_range: - min: 180224 - max: 1670448 + min: 176128 + max: 1522496 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +245,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jmg9o22mg + job_id: jogkkdyvg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +254,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:35:24Z' + timestamp: '2024-08-27T00:48:56Z' - torchscript_onnx_tflite: - inference_time: 309.0 - throughput: 3236.2459546925566 + inference_time: 278.0 + throughput: 3597.122302158273 estimated_peak_memory_range: min: 12288 - max: 1402544 + max: 4433320 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +268,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jw56ommn5 + job_id: j0pxzmy85 job_status: Passed torchscript_onnx_qnn: - inference_time: 338.0 - throughput: 2958.579881656805 + inference_time: 339.0 + throughput: 2949.8525073746314 estimated_peak_memory_range: - min: 188416 - max: 1897584 + min: 172032 + max: 1703808 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +283,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jnp1o11n5 + job_id: jn5qdw2eg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +292,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:35:26Z' + timestamp: '2024-08-27T00:48:57Z' - torchscript_onnx_tflite: - inference_time: 295.0 - throughput: 3389.830508474576 + inference_time: 284.0 + throughput: 3521.1267605633802 estimated_peak_memory_range: - min: 16384 - max: 12951376 + min: 12288 + max: 41387248 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +306,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j1p3o77mp + job_id: jo5ml437g job_status: Passed torchscript_onnx_qnn: - inference_time: 339.0 - throughput: 2949.8525073746314 + inference_time: 337.0 + throughput: 2967.359050445104 estimated_peak_memory_range: - min: 188416 - max: 1688072 + min: 184320 + max: 1670088 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +321,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jvgd6446p + job_id: j1glq7k2p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +330,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:35:27Z' + timestamp: '2024-08-27T00:48:59Z' - torchscript_onnx_tflite: - inference_time: 944.0 - throughput: 1059.322033898305 + inference_time: 880.0 + throughput: 1136.3636363636363 estimated_peak_memory_range: min: 12288 - max: 21201536 + max: 21011920 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +344,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jwgodww15 + job_id: jegnwx3jg job_status: Passed torchscript_onnx_qnn: - inference_time: 1141.0 - throughput: 876.4241893076249 + inference_time: 1170.0 + throughput: 854.7008547008547 estimated_peak_memory_range: min: 12288 - max: 7732160 + max: 8233440 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +359,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jqp4e442g + job_id: j1p3r8mmp job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +368,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:35:29Z' + timestamp: '2024-08-27T00:49:00Z' - torchscript_onnx_tflite: - inference_time: 5803.0 - throughput: 172.32465965879717 + inference_time: 5800.0 + throughput: 172.41379310344828 estimated_peak_memory_range: - min: 12288 - max: 6839264 + min: 24576 + max: 2325864 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +382,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: j1pv2mmzg + job_id: jopr79ekg job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +391,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:35:18Z' + timestamp: '2024-08-27T00:48:52Z' - torchscript_onnx_qnn: - inference_time: 444.0 - throughput: 2252.252252252252 + inference_time: 483.0 + throughput: 2070.3933747412007 estimated_peak_memory_range: - min: 544768 - max: 544768 + min: 1744896 + max: 1744896 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +405,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 86 - job_id: jygzzyy4g + job_id: j2p0xkr0p job_status: Passed torchscript_onnx: - inference_time: 555.0 - throughput: 1801.8018018018017 + inference_time: 541.0 + throughput: 1848.4288354898335 estimated_peak_memory_range: - min: 9515008 - max: 9515008 + min: 9396224 + max: 9396224 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +420,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 91 - job_id: jegn1qqjp + job_id: j7gj81l15 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +429,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:35:33Z' + timestamp: '2024-08-27T00:49:03Z' diff --git a/qai_hub_models/models/hrnet_pose/perf.yaml b/qai_hub_models/models/hrnet_pose/perf.yaml index 38e3bb03..e1c6fdbf 100644 --- a/qai_hub_models/models/hrnet_pose/perf.yaml +++ b/qai_hub_models/models/hrnet_pose/perf.yaml @@ -45,11 +45,11 @@ models: - name: HRNetPose performance_metrics: - torchscript_onnx_tflite: - inference_time: 2867.0 - throughput: 348.7966515521451 + inference_time: 2827.0 + throughput: 353.73187124159887 estimated_peak_memory_range: - min: 32768 - max: 2436936 + min: 24576 + max: 2675032 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 516 - job_id: jogk606v5 + job_id: jnp1m3wnp job_status: Passed torchscript_onnx_qnn: - inference_time: 2994.0 - throughput: 334.001336005344 + inference_time: 2909.0 + throughput: 343.7607425232039 estimated_peak_memory_range: - min: 0 - max: 17123112 + min: 16384 + max: 16687464 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 747 - job_id: j7gj3y31p + job_id: jopr792kg job_status: Passed torchscript_onnx: - inference_time: 3097.0 - throughput: 322.8931223764934 + inference_time: 3067.0 + throughput: 326.0515161395501 estimated_peak_memory_range: - min: 20480 - max: 60487728 + min: 24576 + max: 610876992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 749 - job_id: jqp4e4e2g + job_id: jw560vnn5 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:34:34Z' + timestamp: '2024-08-27T00:48:11Z' - torchscript_onnx_tflite: - inference_time: 2055.0 - throughput: 486.61800486618006 + inference_time: 2033.0 + throughput: 491.88391539596654 estimated_peak_memory_range: min: 16384 - max: 120454832 + max: 122873440 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 516 - job_id: jn5q414eg + job_id: jvgdm0q6g job_status: Passed torchscript_onnx_qnn: - inference_time: 2120.0 - throughput: 471.6981132075472 + inference_time: 2112.0 + throughput: 473.4848484848485 estimated_peak_memory_range: - min: 0 - max: 36162544 + min: 606208 + max: 34402176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 747 - job_id: jlpe6x68g + job_id: jep2zj96p job_status: Passed torchscript_onnx: inference_time: 2217.0 throughput: 451.05999097880016 estimated_peak_memory_range: min: 0 - max: 146429712 + max: 147048832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 749 - job_id: j0px0r08p + job_id: j1p3r8emp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:34:35Z' + timestamp: '2024-08-27T00:48:12Z' - torchscript_onnx_tflite: - inference_time: 2849.0 - throughput: 351.000351000351 + inference_time: 2791.0 + throughput: 358.29451809387314 estimated_peak_memory_range: - min: 24576 - max: 2011704 + min: 28672 + max: 2641800 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 516 - job_id: j1glw8w2p + job_id: jz5786lnp job_status: Passed torchscript_onnx_qnn: - inference_time: 2695.0 - throughput: 371.0575139146568 + inference_time: 2713.0 + throughput: 368.59565057132323 estimated_peak_memory_range: - min: 618496 - max: 1936280 + min: 622592 + max: 1808040 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 747 - job_id: jz5wyzy4g + job_id: j2p0xkl0p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:34:28Z' + timestamp: '2024-08-27T00:48:06Z' - torchscript_onnx_tflite: - inference_time: 3717.0 - throughput: 269.03416733925206 + inference_time: 3735.0 + throughput: 267.7376171352075 estimated_peak_memory_range: - min: 40960 - max: 108743520 + min: 16384 + max: 108533824 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 516 - job_id: jw56omon5 + job_id: jqp428d2g job_status: Passed torchscript_onnx_qnn: - inference_time: 3834.0 - throughput: 260.8242044861763 + inference_time: 3900.0 + throughput: 256.4102564102564 estimated_peak_memory_range: min: 606208 - max: 29126640 + max: 30744720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 747 - job_id: jz57onong + job_id: j1glq732p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:34:33Z' + timestamp: '2024-08-27T00:48:10Z' - torchscript_onnx_tflite: - inference_time: 2842.0 - throughput: 351.8648838845883 + inference_time: 2822.0 + throughput: 354.3586109142452 estimated_peak_memory_range: - min: 229376 - max: 2398944 + min: 225280 + max: 2232792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 516 - job_id: j1p3o7omp + job_id: j0pxzm685 job_status: Passed torchscript_onnx_qnn: - inference_time: 2686.0 - throughput: 372.3008190618019 + inference_time: 2730.0 + throughput: 366.3003663003663 estimated_peak_memory_range: - min: 651264 - max: 2246416 + min: 630784 + max: 2350720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 747 - job_id: jmg9o2omg + job_id: j1p8k8zqp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:34:29Z' + timestamp: '2024-08-27T00:48:07Z' - torchscript_onnx_tflite: - inference_time: 2849.0 - throughput: 351.000351000351 + inference_time: 2827.0 + throughput: 353.73187124159887 estimated_peak_memory_range: - min: 0 - max: 2210568 + min: 16384 + max: 6507128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 516 - job_id: jwgodwd15 + job_id: jo5ml467g job_status: Passed torchscript_onnx_qnn: - inference_time: 2724.0 - throughput: 367.1071953010279 + inference_time: 2744.0 + throughput: 364.4314868804665 estimated_peak_memory_range: - min: 634880 - max: 2236272 + min: 2220032 + max: 3515680 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 747 - job_id: jnp1o1on5 + job_id: jogkkd3vg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:34:31Z' + timestamp: '2024-08-27T00:48:08Z' - torchscript_onnx_tflite: - inference_time: 2818.0 - throughput: 354.86160397444996 + inference_time: 2870.0 + throughput: 348.4320557491289 estimated_peak_memory_range: - min: 32768 - max: 2215024 + min: 241664 + max: 2500656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 516 - job_id: j1pv2m2zg + job_id: jegnwxmjg job_status: Passed torchscript_onnx_qnn: - inference_time: 2745.0 - throughput: 364.29872495446267 + inference_time: 2748.0 + throughput: 363.901018922853 estimated_peak_memory_range: - min: 630784 - max: 2069192 + min: 622592 + max: 1946624 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 747 - job_id: jvgd6466p + job_id: jn5qdw3eg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:34:32Z' + timestamp: '2024-08-27T00:48:09Z' - torchscript_onnx_qnn: - inference_time: 2904.0 - throughput: 344.3526170798898 + inference_time: 2925.0 + throughput: 341.88034188034186 estimated_peak_memory_range: min: 589824 max: 589824 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 747 - job_id: jygzzyz4g + job_id: jqpyynj0p job_status: Passed torchscript_onnx: - inference_time: 2980.0 - throughput: 335.5704697986577 + inference_time: 2941.0 + throughput: 340.02040122407345 estimated_peak_memory_range: - min: 59269120 - max: 59269120 + min: 59224064 + max: 59224064 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 749 - job_id: jo5m9k97g + job_id: jwgo9m31g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:34:36Z' + timestamp: '2024-08-27T00:48:13Z' diff --git a/qai_hub_models/models/hrnet_pose_quantized/perf.yaml b/qai_hub_models/models/hrnet_pose_quantized/perf.yaml index 37d49649..9a119627 100644 --- a/qai_hub_models/models/hrnet_pose_quantized/perf.yaml +++ b/qai_hub_models/models/hrnet_pose_quantized/perf.yaml @@ -48,11 +48,11 @@ models: - name: HRNetPoseQuantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 946.0 - throughput: 1057.0824524312895 + inference_time: 945.0 + throughput: 1058.2010582010582 estimated_peak_memory_range: min: 12288 - max: 1954248 + max: 1817032 primary_compute_unit: NPU precision: int8 layer_info: @@ -60,14 +60,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: jogk602v5 + job_id: j0pxzmnl5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1227.0 - throughput: 814.9959250203749 + inference_time: 1229.0 + throughput: 813.6696501220505 estimated_peak_memory_range: - min: 12288 - max: 20568848 + min: 28672 + max: 22801560 primary_compute_unit: NPU precision: int8 layer_info: @@ -75,7 +75,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: jygzzyl4g + job_id: jn5qdweog job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -84,13 +84,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:33:34Z' + timestamp: '2024-08-27T00:47:18Z' - torchscript_onnx_tflite: - inference_time: 698.0 - throughput: 1432.6647564469913 + inference_time: 705.0 + throughput: 1418.4397163120568 estimated_peak_memory_range: min: 12288 - max: 105288464 + max: 107861872 primary_compute_unit: NPU precision: int8 layer_info: @@ -98,14 +98,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: jn5q41leg + job_id: jo5ml4q9g job_status: Passed torchscript_onnx_qnn: - inference_time: 900.0 - throughput: 1111.111111111111 + inference_time: 905.0 + throughput: 1104.9723756906078 estimated_peak_memory_range: - min: 163840 - max: 31822112 + min: 0 + max: 31659552 primary_compute_unit: NPU precision: int8 layer_info: @@ -113,7 +113,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: jz5wyzl4g + job_id: j1glq72mp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -122,13 +122,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:33:35Z' + timestamp: '2024-08-27T00:47:19Z' - torchscript_onnx_tflite: - inference_time: 938.0 - throughput: 1066.0980810234541 + inference_time: 941.0 + throughput: 1062.6992561105208 estimated_peak_memory_range: min: 12288 - max: 3005672 + max: 1555016 primary_compute_unit: NPU precision: int8 layer_info: @@ -136,14 +136,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: j1glw8y2p + job_id: jegnwxlqg job_status: Passed torchscript_onnx_qnn: - inference_time: 1141.0 - throughput: 876.4241893076249 + inference_time: 1180.0 + throughput: 847.457627118644 estimated_peak_memory_range: min: 172032 - max: 1557704 + max: 1443344 primary_compute_unit: NPU precision: int8 layer_info: @@ -151,7 +151,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: jnp1o1nn5 + job_id: j1p3r81np job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -160,13 +160,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:33:38Z' + timestamp: '2024-08-27T00:47:21Z' - torchscript_onnx_tflite: - inference_time: 1160.0 - throughput: 862.0689655172414 + inference_time: 1177.0 + throughput: 849.6176720475786 estimated_peak_memory_range: - min: 16384 - max: 105445152 + min: 61440 + max: 110253552 primary_compute_unit: NPU precision: int8 layer_info: @@ -174,14 +174,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: jw56om8n5 + job_id: jopr7987g job_status: Passed torchscript_onnx_qnn: - inference_time: 1446.0 - throughput: 691.5629322268327 + inference_time: 1448.0 + throughput: 690.6077348066299 estimated_peak_memory_range: - min: 180224 - max: 32512448 + min: 163840 + max: 35175472 primary_compute_unit: NPU precision: int8 layer_info: @@ -189,7 +189,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: j0px0rl8p + job_id: jlpen2wvp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -198,13 +198,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:33:43Z' + timestamp: '2024-08-27T00:47:25Z' - torchscript_onnx_tflite: - inference_time: 957.0 - throughput: 1044.932079414838 + inference_time: 953.0 + throughput: 1049.3179433368311 estimated_peak_memory_range: - min: 20480 - max: 1922000 + min: 12288 + max: 3377120 primary_compute_unit: NPU precision: int8 layer_info: @@ -212,14 +212,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: j1p3o7zmp + job_id: jep2zj0qp job_status: Passed torchscript_onnx_qnn: - inference_time: 1158.0 - throughput: 863.5578583765113 + inference_time: 1196.0 + throughput: 836.1204013377926 estimated_peak_memory_range: - min: 176128 - max: 1722968 + min: 180224 + max: 1506472 primary_compute_unit: NPU precision: int8 layer_info: @@ -227,7 +227,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: jvgd64d6p + job_id: jwgo9mnkg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -236,13 +236,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:33:39Z' + timestamp: '2024-08-27T00:47:22Z' - torchscript_onnx_tflite: - inference_time: 945.0 - throughput: 1058.2010582010582 + inference_time: 949.0 + throughput: 1053.740779768177 estimated_peak_memory_range: min: 12288 - max: 1711728 + max: 2307536 primary_compute_unit: NPU precision: int8 layer_info: @@ -250,14 +250,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: jwgodwl15 + job_id: jqpyynrlp job_status: Passed torchscript_onnx_qnn: - inference_time: 1159.0 - throughput: 862.8127696289905 + inference_time: 1189.0 + throughput: 841.0428931875525 estimated_peak_memory_range: - min: 180224 - max: 1781904 + min: 184320 + max: 1364184 primary_compute_unit: NPU precision: int8 layer_info: @@ -265,7 +265,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: jz57oneng + job_id: j1pvn4rrg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -274,13 +274,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:33:40Z' + timestamp: '2024-08-27T00:47:23Z' - torchscript_onnx_tflite: - inference_time: 954.0 - throughput: 1048.2180293501049 + inference_time: 946.0 + throughput: 1057.0824524312895 estimated_peak_memory_range: min: 12288 - max: 3529576 + max: 1644048 primary_compute_unit: NPU precision: int8 layer_info: @@ -288,14 +288,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: j1pv2mlzg + job_id: j2p0xk3np job_status: Passed torchscript_onnx_qnn: - inference_time: 1186.0 - throughput: 843.1703204047218 + inference_time: 1169.0 + throughput: 855.4319931565441 estimated_peak_memory_range: - min: 180224 - max: 1692848 + min: 172032 + max: 1765008 primary_compute_unit: NPU precision: int8 layer_info: @@ -303,7 +303,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: jqp4e4y2g + job_id: j7gj812e5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -312,13 +312,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:33:41Z' + timestamp: '2024-08-27T00:47:24Z' - torchscript_onnx_tflite: - inference_time: 3795.0 - throughput: 263.5046113306983 + inference_time: 3883.0 + throughput: 257.53283543651816 estimated_peak_memory_range: - min: 12288 - max: 66857040 + min: 61440 + max: 70986256 primary_compute_unit: NPU precision: int8 layer_info: @@ -326,14 +326,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: j7gj3yr1p + job_id: j1p8k80op job_status: Passed torchscript_onnx_qnn: - inference_time: 5222.0 - throughput: 191.49751053236307 + inference_time: 5426.0 + throughput: 184.29782528566162 estimated_peak_memory_range: - min: 200704 - max: 7978608 + min: 172032 + max: 8441152 primary_compute_unit: NPU precision: int8 layer_info: @@ -341,7 +341,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: jo5m9k07g + job_id: jygz0wjx5 job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -350,13 +350,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:33:44Z' + timestamp: '2024-08-27T00:47:26Z' - torchscript_onnx_tflite: - inference_time: 16894.0 - throughput: 59.192612761927315 + inference_time: 17438.0 + throughput: 57.34602592040372 estimated_peak_memory_range: - min: 106496 - max: 2107272 + min: 102400 + max: 7412016 primary_compute_unit: NPU precision: int8 layer_info: @@ -364,7 +364,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 518 - job_id: jlpe6x78g + job_id: jogkkd7ng job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -373,13 +373,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:33:33Z' + timestamp: '2024-08-27T00:47:17Z' - torchscript_onnx_qnn: - inference_time: 1356.0 - throughput: 737.4631268436578 + inference_time: 1474.0 + throughput: 678.42605156038 estimated_peak_memory_range: - min: 397312 - max: 397312 + min: 1613824 + max: 1613824 primary_compute_unit: NPU precision: int8 layer_info: @@ -387,7 +387,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 487 - job_id: jmg9o2zmg + job_id: jw560vzy5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -396,4 +396,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:33:36Z' + timestamp: '2024-08-27T00:47:20Z' diff --git a/qai_hub_models/models/huggingface_wavlm_base_plus/perf.yaml b/qai_hub_models/models/huggingface_wavlm_base_plus/perf.yaml index ace668e5..75fd6172 100644 --- a/qai_hub_models/models/huggingface_wavlm_base_plus/perf.yaml +++ b/qai_hub_models/models/huggingface_wavlm_base_plus/perf.yaml @@ -45,11 +45,11 @@ models: - name: HuggingFace-WavLM-Base-Plus performance_metrics: - torchscript_onnx_tflite: - inference_time: 835345.0 - throughput: 1.1971101760350513 + inference_time: 922835.0 + throughput: 1.0836173313755981 estimated_peak_memory_range: - min: 149303296 - max: 459427272 + min: 149495808 + max: 152593400 primary_compute_unit: CPU precision: fp32 layer_info: @@ -57,7 +57,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 811 total_layers: 811 - job_id: jvgd643zp + job_id: jn5qdw0og job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -66,13 +66,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:31:03Z' + timestamp: '2024-08-27T00:44:53Z' - torchscript_onnx_tflite: - inference_time: 832045.0 - throughput: 1.201858072580209 + inference_time: 781762.0 + throughput: 1.279161688595762 estimated_peak_memory_range: - min: 149278720 - max: 181724928 + min: 101744640 + max: 134680976 primary_compute_unit: CPU precision: fp32 layer_info: @@ -80,7 +80,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 811 total_layers: 811 - job_id: jz5wyz74g + job_id: j1glq76mp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -89,13 +89,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:31:04Z' + timestamp: '2024-08-27T00:44:54Z' - torchscript_onnx_tflite: - inference_time: 893263.0 - throughput: 1.1194911241146224 + inference_time: 794692.0 + throughput: 1.2583491465876089 estimated_peak_memory_range: - min: 147488768 - max: 150610424 + min: 149463040 + max: 152317696 primary_compute_unit: CPU precision: fp32 layer_info: @@ -103,7 +103,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 811 total_layers: 811 - job_id: jmg9o2mmg + job_id: jw560vey5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -112,13 +112,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:31:06Z' + timestamp: '2024-08-27T00:44:55Z' - torchscript_onnx_tflite: - inference_time: 1196696.0 - throughput: 0.8356341125899978 + inference_time: 1234599.0 + throughput: 0.8099795966139612 estimated_peak_memory_range: - min: 99954688 - max: 136088496 + min: 94580736 + max: 128826992 primary_compute_unit: CPU precision: fp32 layer_info: @@ -126,7 +126,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 811 total_layers: 811 - job_id: jnp1o1jn5 + job_id: j1p3r8vnp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -135,13 +135,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:31:07Z' + timestamp: '2024-08-27T00:44:56Z' - torchscript_onnx_tflite: - inference_time: 857063.0 - throughput: 1.1667753712387536 + inference_time: 988047.0 + throughput: 1.0120976026444086 estimated_peak_memory_range: - min: 149209088 - max: 634902648 + min: 145399808 + max: 149393776 primary_compute_unit: CPU precision: fp32 layer_info: @@ -149,7 +149,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 811 total_layers: 811 - job_id: jvgd6436p + job_id: jwgo9mkkg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -158,13 +158,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:31:08Z' + timestamp: '2024-08-27T00:44:57Z' - torchscript_onnx_tflite: - inference_time: 1003917.0 - throughput: 0.9960982830253895 + inference_time: 940375.0 + throughput: 1.0634055562940317 estimated_peak_memory_range: - min: 149295104 - max: 151872744 + min: 147849216 + max: 174405384 primary_compute_unit: CPU precision: fp32 layer_info: @@ -172,7 +172,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 811 total_layers: 811 - job_id: jz57on4ng + job_id: j1pvn40rg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -181,13 +181,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:31:09Z' + timestamp: '2024-08-27T00:44:58Z' - torchscript_onnx_tflite: - inference_time: 829986.0 - throughput: 1.2048395997040915 + inference_time: 864077.0 + throughput: 1.15730426802241 estimated_peak_memory_range: - min: 149540864 - max: 152494712 + min: 148320256 + max: 154912536 primary_compute_unit: CPU precision: fp32 layer_info: @@ -195,7 +195,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 811 total_layers: 811 - job_id: jqp4e412g + job_id: j7gj81ze5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -204,4 +204,4 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:31:10Z' + timestamp: '2024-08-27T00:44:59Z' diff --git a/qai_hub_models/models/inception_v3/perf.yaml b/qai_hub_models/models/inception_v3/perf.yaml index 9ebc72d8..aae2a0e0 100644 --- a/qai_hub_models/models/inception_v3/perf.yaml +++ b/qai_hub_models/models/inception_v3/perf.yaml @@ -45,11 +45,11 @@ models: - name: Inception-v3 performance_metrics: - torchscript_onnx_tflite: - inference_time: 1358.0 - throughput: 736.3770250368188 + inference_time: 1329.0 + throughput: 752.4454477050414 estimated_peak_memory_range: - min: 12288 - max: 53261352 + min: 0 + max: 2506112 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: jqp4e431g + job_id: j1glq74mp job_status: Passed torchscript_onnx_qnn: - inference_time: 1417.0 - throughput: 705.7163020465773 + inference_time: 1398.0 + throughput: 715.307582260372 estimated_peak_memory_range: - min: 622592 - max: 149841664 + min: 0 + max: 149432368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: j2p0o96np + job_id: jygz0w2x5 job_status: Passed torchscript_onnx: - inference_time: 1722.0 - throughput: 580.7200929152149 + inference_time: 1718.0 + throughput: 582.0721769499418 estimated_peak_memory_range: - min: 12288 - max: 52282080 + min: 16384 + max: 785071008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 221 - job_id: j1pv2mdrg + job_id: jegnwx0qg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:30:32Z' + timestamp: '2024-08-27T00:44:23Z' - torchscript_onnx_tflite: - inference_time: 1008.0 - throughput: 992.063492063492 + inference_time: 1004.0 + throughput: 996.01593625498 estimated_peak_memory_range: min: 16384 - max: 59057744 + max: 59531456 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: j0px0rxlp + job_id: jw560v2y5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1046.0 - throughput: 956.0229445506692 + inference_time: 1045.0 + throughput: 956.9377990430622 estimated_peak_memory_range: min: 0 - max: 19460944 + max: 18294928 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: j1p8jr1o5 + job_id: jz5wrxwmp job_status: Passed torchscript_onnx: - inference_time: 1297.0 - throughput: 771.0100231303007 + inference_time: 1276.0 + throughput: 783.6990595611285 estimated_peak_memory_range: min: 0 - max: 56692528 + max: 57810720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 221 - job_id: j7gj3y7ep + job_id: jopr7967g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:30:33Z' + timestamp: '2024-08-27T00:44:24Z' - torchscript_onnx_tflite: - inference_time: 1346.0 - throughput: 742.9420505200594 + inference_time: 1324.0 + throughput: 755.2870090634441 estimated_peak_memory_range: - min: 0 - max: 597336688 + min: 20480 + max: 608201480 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: jo5m9k89g + job_id: j1p3r8nnp job_status: Passed torchscript_onnx_qnn: - inference_time: 1428.0 - throughput: 700.2801120448179 + inference_time: 1422.0 + throughput: 703.2348804500704 estimated_peak_memory_range: - min: 634880 - max: 2260736 + min: 638976 + max: 2303752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jn5q41vog + job_id: jvgdm0nzg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:30:26Z' + timestamp: '2024-08-27T00:44:19Z' - torchscript_onnx_tflite: - inference_time: 2165.0 - throughput: 461.8937644341801 + inference_time: 2101.0 + throughput: 475.9638267491671 estimated_peak_memory_range: - min: 16384 - max: 60001792 + min: 20480 + max: 60300288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: jegn1qkqp + job_id: jwgo9mzkg job_status: Passed torchscript_onnx_qnn: - inference_time: 2205.0 - throughput: 453.51473922902494 + inference_time: 2183.0 + throughput: 458.0852038479157 estimated_peak_memory_range: - min: 528384 - max: 20943712 + min: 626688 + max: 20371168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jwgodwrk5 + job_id: jo5ml4e9g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:30:31Z' + timestamp: '2024-08-27T00:44:22Z' - torchscript_onnx_tflite: - inference_time: 1364.0 - throughput: 733.1378299120234 + inference_time: 1324.0 + throughput: 755.2870090634441 estimated_peak_memory_range: - min: 49152 - max: 1626768 + min: 16384 + max: 2232504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: joprxdw7p + job_id: j1pvn4qrg job_status: Passed torchscript_onnx_qnn: - inference_time: 1442.0 - throughput: 693.4812760055479 + inference_time: 1440.0 + throughput: 694.4444444444445 estimated_peak_memory_range: min: 638976 - max: 1893816 + max: 1893496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: j1glw8xmp + job_id: jz578629p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:30:27Z' + timestamp: '2024-08-27T00:44:20Z' - torchscript_onnx_tflite: - inference_time: 1359.0 - throughput: 735.8351729212657 + inference_time: 1324.0 + throughput: 755.2870090634441 estimated_peak_memory_range: - min: 28672 - max: 2082520 + min: 40960 + max: 31010416 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: jep2odeqg + job_id: j7gj81de5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1441.0 - throughput: 693.9625260235947 + inference_time: 1430.0 + throughput: 699.3006993006993 estimated_peak_memory_range: - min: 626688 - max: 2232656 + min: 647168 + max: 2005080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jw56om7y5 + job_id: jqp428n1g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:30:28Z' + timestamp: '2024-08-27T00:44:21Z' - torchscript_onnx_tflite: - inference_time: 1361.0 - throughput: 734.7538574577517 + inference_time: 1327.0 + throughput: 753.5795026375282 estimated_peak_memory_range: - min: 36864 - max: 2114488 + min: 32768 + max: 598225768 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 129 - job_id: jqpy82mlg + job_id: jlpen2ovp job_status: Passed torchscript_onnx_qnn: - inference_time: 1416.0 - throughput: 706.2146892655368 + inference_time: 1417.0 + throughput: 705.7163020465773 estimated_peak_memory_range: - min: 643072 - max: 2294848 + min: 647168 + max: 2022696 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: j1p3o79np + job_id: j0pxzm9l5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:30:29Z' + timestamp: '2024-08-27T00:44:22Z' - torchscript_onnx_qnn: - inference_time: 1437.0 - throughput: 695.8942240779402 + inference_time: 1468.0 + throughput: 681.1989100817439 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jogk608n5 + job_id: jmg9q808p job_status: Passed torchscript_onnx: - inference_time: 1676.0 - throughput: 596.6587112171837 + inference_time: 1713.0 + throughput: 583.7711617046118 estimated_peak_memory_range: - min: 49856512 - max: 49856512 + min: 50036736 + max: 50036736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 221 - job_id: jlpe6xzvg + job_id: jep2zjxqp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:30:34Z' + timestamp: '2024-08-27T00:44:25Z' diff --git a/qai_hub_models/models/inception_v3_quantized/perf.yaml b/qai_hub_models/models/inception_v3_quantized/perf.yaml index 62cd4cbe..b5086985 100644 --- a/qai_hub_models/models/inception_v3_quantized/perf.yaml +++ b/qai_hub_models/models/inception_v3_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: Inception-v3-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 621.0 - throughput: 1610.3059581320451 + inference_time: 588.0 + throughput: 1700.6802721088436 estimated_peak_memory_range: - min: 12288 - max: 211519592 + min: 28672 + max: 205288832 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jz57on79g + job_id: jqpyyn3lp job_status: Passed torchscript_onnx_qnn: - inference_time: 647.0 - throughput: 1545.595054095827 + inference_time: 646.0 + throughput: 1547.9876160990711 estimated_peak_memory_range: - min: 12288 - max: 249300024 + min: 32768 + max: 251250176 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: j1p8jr4o5 + job_id: j1pvn46rg job_status: Passed torchscript_onnx: - inference_time: 852.0 - throughput: 1173.7089201877934 + inference_time: 857.0 + throughput: 1166.8611435239206 estimated_peak_memory_range: min: 12288 - max: 31009584 + max: 30978064 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: jlpe6xyvg + job_id: jqp428l1g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:29:49Z' + timestamp: '2024-08-27T00:43:46Z' - torchscript_onnx_tflite: - inference_time: 449.0 - throughput: 2227.1714922048996 + inference_time: 453.0 + throughput: 2207.5055187637968 estimated_peak_memory_range: min: 12288 - max: 70494560 + max: 73568400 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jqp4e491g + job_id: j2p0xk0np job_status: Passed torchscript_onnx_qnn: - inference_time: 504.0 - throughput: 1984.126984126984 + inference_time: 495.0 + throughput: 2020.20202020202 estimated_peak_memory_range: - min: 0 - max: 16692720 + min: 167936 + max: 20641696 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: jogk609n5 + job_id: j7gj81ve5 job_status: Passed torchscript_onnx: - inference_time: 685.0 - throughput: 1459.85401459854 + inference_time: 650.0 + throughput: 1538.4615384615386 estimated_peak_memory_range: - min: 0 - max: 96811840 + min: 12288 + max: 98343200 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: jygzzynxg + job_id: j0pxzmkl5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:29:50Z' + timestamp: '2024-08-27T00:43:47Z' - torchscript_onnx_tflite: - inference_time: 609.0 - throughput: 1642.0361247947455 + inference_time: 583.0 + throughput: 1715.2658662092624 estimated_peak_memory_range: - min: 12288 - max: 14360920 + min: 45056 + max: 215658216 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: j0px0rdlp + job_id: j1p8k8yop job_status: Passed torchscript_onnx_qnn: - inference_time: 663.0 - throughput: 1508.2956259426849 + inference_time: 667.0 + throughput: 1499.2503748125937 estimated_peak_memory_range: min: 184320 - max: 1353912 + max: 1743168 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: j1glw8lmp + job_id: jygz0w3x5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:29:42Z' + timestamp: '2024-08-27T00:43:40Z' - torchscript_onnx_tflite: - inference_time: 729.0 - throughput: 1371.7421124828531 + inference_time: 706.0 + throughput: 1416.4305949008499 estimated_peak_memory_range: - min: 20480 - max: 72330704 + min: 16384 + max: 73929104 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jo5m9kd9g + job_id: jogkkdxng job_status: Passed torchscript_onnx_qnn: - inference_time: 782.0 - throughput: 1278.772378516624 + inference_time: 761.0 + throughput: 1314.060446780552 estimated_peak_memory_range: min: 167936 - max: 20279584 + max: 20963232 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: j1pv2m7rg + job_id: jvgdm0xzg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:29:47Z' + timestamp: '2024-08-27T00:43:44Z' - torchscript_onnx_tflite: - inference_time: 621.0 - throughput: 1610.3059581320451 + inference_time: 585.0 + throughput: 1709.4017094017095 estimated_peak_memory_range: - min: 16384 - max: 6398936 + min: 32768 + max: 205272824 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jegn1q7qp + job_id: jn5qdwqog job_status: Passed torchscript_onnx_qnn: - inference_time: 670.0 - throughput: 1492.5373134328358 + inference_time: 651.0 + throughput: 1536.0983102918588 estimated_peak_memory_range: - min: 184320 - max: 1640224 + min: 180224 + max: 1507744 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: jw56omwy5 + job_id: jz5wrxemp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:29:43Z' + timestamp: '2024-08-27T00:43:41Z' - torchscript_onnx_tflite: - inference_time: 614.0 - throughput: 1628.6644951140065 + inference_time: 583.0 + throughput: 1715.2658662092624 estimated_peak_memory_range: - min: 40960 - max: 27914496 + min: 12288 + max: 10660824 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: joprxdn7p + job_id: j1glq7vmp job_status: Passed torchscript_onnx_qnn: - inference_time: 657.0 - throughput: 1522.0700152207 + inference_time: 656.0 + throughput: 1524.3902439024391 estimated_peak_memory_range: min: 184320 - max: 1471096 + max: 1515416 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: j1p3o76np + job_id: jmg9q8l8p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:29:44Z' + timestamp: '2024-08-27T00:43:42Z' - torchscript_onnx_tflite: - inference_time: 602.0 - throughput: 1661.1295681063123 + inference_time: 586.0 + throughput: 1706.4846416382252 estimated_peak_memory_range: - min: 28672 - max: 264524200 + min: 12288 + max: 1487280 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jep2odvqg + job_id: jw560vyy5 job_status: Passed torchscript_onnx_qnn: - inference_time: 666.0 - throughput: 1501.5015015015015 + inference_time: 649.0 + throughput: 1540.8320493066255 estimated_peak_memory_range: - min: 217088 - max: 1493664 + min: 188416 + max: 1485096 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: jwgodw8k5 + job_id: jnp1m347p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:29:46Z' + timestamp: '2024-08-27T00:43:43Z' - torchscript_onnx_tflite: - inference_time: 2362.0 - throughput: 423.3700254022015 + inference_time: 2462.0 + throughput: 406.17384240454913 estimated_peak_memory_range: - min: 12288 - max: 26824576 + min: 16384 + max: 27787168 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jqpy827lg + job_id: j1p3r8jnp job_status: Passed torchscript_onnx_qnn: - inference_time: 2853.0 - throughput: 350.5082369435682 + inference_time: 3018.0 + throughput: 331.3452617627568 estimated_peak_memory_range: min: 12288 - max: 7483936 + max: 7864960 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: j7gj3yqep + job_id: jz5786y9p job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:29:48Z' + timestamp: '2024-08-27T00:43:45Z' - torchscript_onnx_tflite: - inference_time: 7762.0 - throughput: 128.83277505797474 + inference_time: 7770.0 + throughput: 128.7001287001287 estimated_peak_memory_range: - min: 45056 - max: 1967496 + min: 12288 + max: 2073272 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: j2p0o9vnp + job_id: jwgo9m2kg job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:29:37Z' + timestamp: '2024-08-27T00:43:36Z' - torchscript_onnx_qnn: - inference_time: 726.0 - throughput: 1377.4104683195592 + inference_time: 707.0 + throughput: 1414.4271570014143 estimated_peak_memory_range: - min: 495616 - max: 495616 + min: 503808 + max: 503808 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: jn5q41mog + job_id: jlpen2dvp job_status: Passed torchscript_onnx: - inference_time: 810.0 - throughput: 1234.567901234568 + inference_time: 813.0 + throughput: 1230.0123001230013 estimated_peak_memory_range: - min: 28663808 - max: 28663808 + min: 28508160 + max: 28508160 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 134 - job_id: jz5wyz4mg + job_id: jo5ml4n9g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:29:51Z' + timestamp: '2024-08-27T00:43:48Z' diff --git a/qai_hub_models/models/lama_dilated/perf.yaml b/qai_hub_models/models/lama_dilated/perf.yaml index 9a11fcae..7d505bb4 100644 --- a/qai_hub_models/models/lama_dilated/perf.yaml +++ b/qai_hub_models/models/lama_dilated/perf.yaml @@ -45,11 +45,11 @@ models: - name: LaMa-Dilated performance_metrics: - torchscript_onnx_tflite: - inference_time: 75260.0 - throughput: 13.287270794578793 + inference_time: 76184.0 + throughput: 13.126115719836186 estimated_peak_memory_range: - min: 3268608 - max: 138460800 + min: 3284992 + max: 138432896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 343 - job_id: jo5m9ko9g + job_id: j1glqerjp job_status: Passed torchscript_onnx_qnn: - inference_time: 71112.0 - throughput: 14.062324220947238 + inference_time: 70540.0 + throughput: 14.176353841791892 estimated_peak_memory_range: - min: 1150976 - max: 44966040 + min: 3194880 + max: 35134240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 333 - job_id: jogk60on5 + job_id: jygz0wdo5 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,13 +81,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:28:23Z' + timestamp: '2024-08-27T00:42:26Z' - torchscript_onnx_tflite: - inference_time: 51490.0 - throughput: 19.42124684404739 + inference_time: 51200.0 + throughput: 19.53125 estimated_peak_memory_range: - min: 3215360 - max: 239320304 + min: 2404352 + max: 239766048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -95,14 +95,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 343 - job_id: jegn1qoqp + job_id: jw560ql65 job_status: Passed torchscript_onnx_qnn: - inference_time: 49768.0 - throughput: 20.09323259926057 + inference_time: 48295.0 + throughput: 20.706077233668083 estimated_peak_memory_range: - min: 4239360 - max: 88505392 + min: 4214784 + max: 87933648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 333 - job_id: jn5q41zog + job_id: jz5wrx63p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,13 +119,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:28:24Z' + timestamp: '2024-08-27T00:42:27Z' - torchscript_onnx_tflite: - inference_time: 75057.0 - throughput: 13.323207695484765 + inference_time: 74716.0 + throughput: 13.384014133518924 estimated_peak_memory_range: - min: 24576 - max: 302919272 + min: 2195456 + max: 138103720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -133,14 +133,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 343 - job_id: joprxdo7p + job_id: j1p3rq23p job_status: Passed torchscript_onnx_qnn: - inference_time: 67711.0 - throughput: 14.768649111665756 + inference_time: 66694.0 + throughput: 14.993852520466609 estimated_peak_memory_range: - min: 4399104 - max: 5657352 + min: 4374528 + max: 5458784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -148,7 +148,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 333 - job_id: jw56omdy5 + job_id: jnp1m3z8p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -157,13 +157,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:28:27Z' + timestamp: '2024-08-27T00:42:29Z' - torchscript_onnx_tflite: - inference_time: 105020.0 - throughput: 9.521995810321844 + inference_time: 105579.0 + throughput: 9.471580522641814 estimated_peak_memory_range: - min: 3411968 - max: 150842160 + min: 3239936 + max: 151910944 primary_compute_unit: NPU precision: fp16 layer_info: @@ -171,14 +171,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 343 - job_id: jep2od4qg + job_id: jwgo9eqqg job_status: Passed torchscript_onnx_qnn: - inference_time: 98933.0 - throughput: 10.107850767691266 + inference_time: 102549.0 + throughput: 9.75143589893612 estimated_peak_memory_range: min: 4214784 - max: 43962352 + max: 45514192 primary_compute_unit: NPU precision: fp16 layer_info: @@ -186,7 +186,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 333 - job_id: j7gj3ywep + job_id: jnp1m3z7p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -195,13 +195,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:28:32Z' + timestamp: '2024-08-27T00:42:33Z' - torchscript_onnx_tflite: - inference_time: 75182.0 - throughput: 13.301056103854647 + inference_time: 74836.0 + throughput: 13.362552782083489 estimated_peak_memory_range: - min: 3301376 - max: 54154832 + min: 3248128 + max: 138401072 primary_compute_unit: NPU precision: fp16 layer_info: @@ -209,14 +209,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 343 - job_id: jqpy82qlg + job_id: j1pvn4kkg job_status: Passed torchscript_onnx_qnn: - inference_time: 67404.0 - throughput: 14.835914782505489 + inference_time: 66927.0 + throughput: 14.941652845637785 estimated_peak_memory_range: - min: 4280320 - max: 5676072 + min: 6160384 + max: 7774928 primary_compute_unit: NPU precision: fp16 layer_info: @@ -224,7 +224,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 333 - job_id: j1p3o7wnp + job_id: jvgdm01rg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -233,13 +233,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:28:28Z' + timestamp: '2024-08-27T00:42:30Z' - torchscript_onnx_tflite: - inference_time: 75103.0 - throughput: 13.315047334993276 + inference_time: 74964.0 + throughput: 13.339736406808601 estimated_peak_memory_range: - min: 3276800 - max: 138377136 + min: 3186688 + max: 347782248 primary_compute_unit: NPU precision: fp16 layer_info: @@ -247,14 +247,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 343 - job_id: j2p0o9dnp + job_id: j7gj81nv5 job_status: Passed torchscript_onnx_qnn: - inference_time: 67508.0 - throughput: 14.813059192984536 + inference_time: 67161.0 + throughput: 14.889593662988936 estimated_peak_memory_range: - min: 4378624 - max: 5613544 + min: 4395008 + max: 5509352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -262,7 +262,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 333 - job_id: jwgodw4k5 + job_id: jz5wrx6mp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -271,13 +271,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:28:29Z' + timestamp: '2024-08-27T00:42:31Z' - torchscript_onnx_tflite: - inference_time: 75149.0 - throughput: 13.306896964696802 + inference_time: 74902.0 + throughput: 13.350778350377826 estimated_peak_memory_range: - min: 3284992 - max: 138502168 + min: 3289088 + max: 221792512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -285,14 +285,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 343 - job_id: j1p8jr6o5 + job_id: jlpen2mop job_status: Passed torchscript_onnx_qnn: - inference_time: 67152.0 - throughput: 14.89158923040267 + inference_time: 66280.0 + throughput: 15.087507543753771 estimated_peak_memory_range: - min: 4415488 - max: 5614120 + min: 4378624 + max: 5642632 primary_compute_unit: NPU precision: fp16 layer_info: @@ -300,7 +300,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 333 - job_id: j1pv2m9rg + job_id: jmg9q8n8p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -309,10 +309,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:28:30Z' + timestamp: '2024-08-27T00:42:32Z' - torchscript_onnx_qnn: - inference_time: 70578.0 - throughput: 14.168721131230695 + inference_time: 69247.0 + throughput: 14.441058818432568 estimated_peak_memory_range: min: 4202496 max: 4202496 @@ -323,7 +323,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 333 - job_id: j1glw81mp + job_id: jmg9q8nwp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -332,4 +332,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:28:26Z' + timestamp: '2024-08-27T00:42:28Z' diff --git a/qai_hub_models/models/litehrnet/perf.yaml b/qai_hub_models/models/litehrnet/perf.yaml index a088bdaf..ae216fc3 100644 --- a/qai_hub_models/models/litehrnet/perf.yaml +++ b/qai_hub_models/models/litehrnet/perf.yaml @@ -45,42 +45,19 @@ models: - name: LiteHRNet performance_metrics: - torchscript_onnx_tflite: - inference_time: 11098.0 - throughput: 90.10632546404757 + inference_time: 4866.0 + throughput: 205.5076037813399 estimated_peak_memory_range: - min: 6332416 - max: 197843136 + min: 233472 + max: 89151264 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1225 + layers_on_npu: 1233 layers_on_gpu: 0 - layers_on_cpu: 10 + layers_on_cpu: 2 total_layers: 1235 - job_id: j1glw8omp - job_status: Passed - reference_device_info: - name: Samsung Galaxy S23 - os: '13' - form_factor: Phone - os_name: Android - manufacturer: Samsung - chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:27:27Z' - - torchscript_onnx_tflite: - inference_time: 7609.0 - throughput: 131.42331449599158 - estimated_peak_memory_range: - min: 6553600 - max: 90714208 - primary_compute_unit: NPU - precision: fp16 - layer_info: - layers_on_npu: 1225 - layers_on_gpu: 0 - layers_on_cpu: 10 - total_layers: 1235 - job_id: jw56omry5 + job_id: jnp1mqx8p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -89,21 +66,21 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:27:28Z' + timestamp: '2024-08-27T00:41:37Z' - torchscript_onnx_tflite: - inference_time: 11179.0 - throughput: 89.45343948474819 + inference_time: 7859.0 + throughput: 127.2426517368622 estimated_peak_memory_range: - min: 6529024 - max: 219391872 + min: 274432 + max: 2044520 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1225 + layers_on_npu: 1233 layers_on_gpu: 0 - layers_on_cpu: 10 + layers_on_cpu: 2 total_layers: 1235 - job_id: j1p3o7xnp + job_id: jvgdm7lrg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -112,21 +89,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:27:29Z' + timestamp: '2024-08-27T00:41:38Z' - torchscript_onnx_tflite: - inference_time: 13355.0 - throughput: 74.87832272557095 + inference_time: 8571.0 + throughput: 116.67250029168125 estimated_peak_memory_range: - min: 6529024 - max: 84309296 + min: 253952 + max: 81461328 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1225 + layers_on_npu: 1233 layers_on_gpu: 0 - layers_on_cpu: 10 + layers_on_cpu: 2 total_layers: 1235 - job_id: jwgodwok5 + job_id: jz578v3vp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -135,21 +112,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:27:31Z' + timestamp: '2024-08-27T00:41:39Z' - torchscript_onnx_tflite: - inference_time: 11202.0 - throughput: 89.26977325477593 + inference_time: 7907.0 + throughput: 126.4702162640698 estimated_peak_memory_range: - min: 6529024 - max: 18616776 + min: 262144 + max: 155993552 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1225 + layers_on_npu: 1233 layers_on_gpu: 0 - layers_on_cpu: 10 + layers_on_cpu: 2 total_layers: 1235 - job_id: j1pv2merg + job_id: jqp42j08g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -158,21 +135,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:27:32Z' + timestamp: '2024-08-27T00:41:39Z' - torchscript_onnx_tflite: - inference_time: 11272.0 - throughput: 88.71540099361249 + inference_time: 7901.0 + throughput: 126.56625743576762 estimated_peak_memory_range: - min: 6529024 - max: 9064952 + min: 249856 + max: 158517416 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1225 + layers_on_npu: 1233 layers_on_gpu: 0 - layers_on_cpu: 10 + layers_on_cpu: 2 total_layers: 1235 - job_id: j7gj3yoep + job_id: j0pxze235 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -181,21 +158,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:27:33Z' + timestamp: '2024-08-27T00:41:40Z' - torchscript_onnx_tflite: - inference_time: 11212.0 - throughput: 89.19015340706386 + inference_time: 7862.0 + throughput: 127.1940981938438 estimated_peak_memory_range: - min: 6537216 - max: 9317792 + min: 258048 + max: 2295352 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1225 + layers_on_npu: 1233 layers_on_gpu: 0 - layers_on_cpu: 10 + layers_on_cpu: 2 total_layers: 1235 - job_id: jlpe6x8vg + job_id: jo5mlvydg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -204,4 +181,4 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:27:34Z' + timestamp: '2024-08-27T00:41:42Z' diff --git a/qai_hub_models/models/llama_v2_7b_chat_quantized/export.py b/qai_hub_models/models/llama_v2_7b_chat_quantized/export.py index 886503a6..aff761ba 100644 --- a/qai_hub_models/models/llama_v2_7b_chat_quantized/export.py +++ b/qai_hub_models/models/llama_v2_7b_chat_quantized/export.py @@ -136,7 +136,7 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} profile_options_per_component: Dict[str, str] = {} - for i, component_name in enumerate(components): + for component_name in components: # Load model part component = model.load_model_part(component_name) @@ -145,7 +145,6 @@ def export_model( **get_input_spec_kwargs(component, additional_model_kwargs) ) - # Trace the model source_model = component.convert_to_hub_source_model( target_runtime, output_path, diff --git a/qai_hub_models/models/llama_v2_7b_chat_quantized/model.py b/qai_hub_models/models/llama_v2_7b_chat_quantized/model.py index 7e97997f..9d3d5783 100644 --- a/qai_hub_models/models/llama_v2_7b_chat_quantized/model.py +++ b/qai_hub_models/models/llama_v2_7b_chat_quantized/model.py @@ -47,7 +47,7 @@ MODEL_ID = __name__.split(".")[-2] -MODEL_ASSET_VERSION = 6 +MODEL_ASSET_VERSION = 7 # Configs AIMET_ENCODINGS_PREFIX = "config" @@ -94,6 +94,13 @@ DEFAULT_USER_PROMPT = "Hi! What is 2+3?" +def _get_intermediate_logit_name(split_part): + if split_part == 1: + return "input_ids" + start_layer_num = MODEL_SPLIT_MAP[split_part][0] + return f"layers_{start_layer_num-1}_add_out_0" + + def get_input_prompt_with_tags( previous_history: str = "", system_context_prompt: str = DEFAULT_PROMPT_CONTEXT, @@ -173,6 +180,7 @@ def __init__( config.use_conv = True config.mask_neg = -100 config.split_model = split_part + if split_part < 1 or split_part > 4: raise RuntimeError( f"Llama2 split_part must be within 1-4 (Provided {split_part})." @@ -395,6 +403,17 @@ def get_input_spec( "position_ids_sin": ((1, 1, input_seq_length, POS_EMBED_DIM), "float32"), } + @staticmethod + def get_output_names(): + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=1, model_split_map=MODEL_SPLIT_MAP + ) + return Llama_QuantizedMixin.get_output_names( + start=layers_start, + end=layers_end, + past_key_val_heads=NUM_KEY_VAL_HEADS, + ) + @staticmethod def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): data = load_input_cached_data( @@ -493,12 +512,26 @@ def get_input_spec( # This can be used with the qai_hub python API to declare # the model input specification upon submitting a compile job. return { - "input_ids": ((1, input_seq_length, ATTENTION_HIDDEN_DIM), "float32"), + _get_intermediate_logit_name(split_part=2): ( + (1, input_seq_length, ATTENTION_HIDDEN_DIM), + "float32", + ), "attention_mask": ((1, 1, input_seq_length, input_seq_length), "float32"), "position_ids_cos": ((1, 1, input_seq_length, POS_EMBED_DIM), "float32"), "position_ids_sin": ((1, 1, input_seq_length, POS_EMBED_DIM), "float32"), } + @staticmethod + def get_output_names(): + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=2, model_split_map=MODEL_SPLIT_MAP + ) + return Llama_QuantizedMixin.get_output_names( + start=layers_start, + end=layers_end, + past_key_val_heads=NUM_KEY_VAL_HEADS, + ) + @staticmethod def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): data = load_input_cached_data( @@ -520,7 +553,7 @@ def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): del model new_inputs = {} - new_inputs["input_ids"] = output[0].detach() + new_inputs[_get_intermediate_logit_name(split_part=2)] = output[0].detach() new_inputs["attention_mask"] = inputs["attention_mask"] new_inputs["position_ids_cos"] = inputs["position_ids_cos"] new_inputs["position_ids_sin"] = inputs["position_ids_sin"] @@ -584,12 +617,26 @@ def get_input_spec( # This can be used with the qai_hub python API to declare # the model input specification upon submitting a compile job. return { - "input_ids": ((1, input_seq_length, ATTENTION_HIDDEN_DIM), "float32"), + _get_intermediate_logit_name(split_part=3): ( + (1, input_seq_length, ATTENTION_HIDDEN_DIM), + "float32", + ), "attention_mask": ((1, 1, input_seq_length, input_seq_length), "float32"), "position_ids_cos": ((1, 1, input_seq_length, POS_EMBED_DIM), "float32"), "position_ids_sin": ((1, 1, input_seq_length, POS_EMBED_DIM), "float32"), } + @staticmethod + def get_output_names(): + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=3, model_split_map=MODEL_SPLIT_MAP + ) + return Llama_QuantizedMixin.get_output_names( + start=layers_start, + end=layers_end, + past_key_val_heads=NUM_KEY_VAL_HEADS, + ) + @staticmethod def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): data = load_input_cached_data( @@ -611,7 +658,7 @@ def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): del model new_inputs = {} - new_inputs["input_ids"] = output[0].detach() + new_inputs[_get_intermediate_logit_name(split_part=3)] = output[0].detach() new_inputs["attention_mask"] = inputs["attention_mask"] new_inputs["position_ids_cos"] = inputs["position_ids_cos"] new_inputs["position_ids_sin"] = inputs["position_ids_sin"] @@ -675,7 +722,10 @@ def get_input_spec( # This can be used with the qai_hub python API to declare # the model input specification upon submitting a compile job. return { - "input_ids": ((1, input_seq_length, ATTENTION_HIDDEN_DIM), "float32"), + _get_intermediate_logit_name(split_part=4): ( + (1, input_seq_length, ATTENTION_HIDDEN_DIM), + "float32", + ), "attention_mask": ((1, 1, input_seq_length, input_seq_length), "float32"), "position_ids_cos": ((1, 1, input_seq_length, POS_EMBED_DIM), "float32"), "position_ids_sin": ((1, 1, input_seq_length, POS_EMBED_DIM), "float32"), @@ -713,7 +763,7 @@ def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): output = model(*inputs.values()) new_inputs = {} - new_inputs["input_ids"] = output[0].detach() + new_inputs[_get_intermediate_logit_name(split_part=4)] = output[0].detach() new_inputs["attention_mask"] = inputs["attention_mask"] new_inputs["position_ids_cos"] = inputs["position_ids_cos"] new_inputs["position_ids_sin"] = inputs["position_ids_sin"] @@ -799,7 +849,15 @@ def get_input_spec( } # Collect past_key_values and drop output names - past_key_val_names = get_past_key_names() + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=1, model_split_map=MODEL_SPLIT_MAP + ) + past_key_val_names = get_past_key_names( + start=layers_start, + end=layers_end, + num_of_past_key_heads=NUM_KEY_VAL_HEADS, + suffix="_in", + ) for past_key_val in past_key_val_names: if "key" in past_key_val: input_spec[past_key_val] = ( @@ -813,6 +871,17 @@ def get_input_spec( ) return input_spec + @staticmethod + def get_output_names(): + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=1, model_split_map=MODEL_SPLIT_MAP + ) + return Llama_QuantizedMixin.get_output_names( + start=layers_start, + end=layers_end, + past_key_val_heads=NUM_KEY_VAL_HEADS, + ) + @staticmethod def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): data = load_input_cached_data( @@ -879,7 +948,12 @@ def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): "position_ids_sin": position_ids_sin, } - key_val = get_past_keyval_with_shift(output[1:], NUM_KEY_VAL_HEADS) + layers_start, _ = get_hidden_layer_range_from_split( + split_part=1, model_split_map=MODEL_SPLIT_MAP + ) + key_val = get_past_keyval_with_shift( + output[1:], layers_start, NUM_KEY_VAL_HEADS, new_key_suffix="_in" + ) for key, val in key_val.items(): data[key] = val @@ -955,14 +1029,25 @@ def get_input_spec( # the model input specification upon submitting a compile job. input_spec = { - "input_ids": ((1, 1, ATTENTION_HIDDEN_DIM), "float32"), + _get_intermediate_logit_name(split_part=2): ( + (1, 1, ATTENTION_HIDDEN_DIM), + "float32", + ), "attention_mask": ((1, 1, 1, input_seq_length), "float32"), "position_ids_cos": ((1, 1, 1, POS_EMBED_DIM), "float32"), "position_ids_sin": ((1, 1, 1, POS_EMBED_DIM), "float32"), } # Collect past_key_values and drop output names - past_key_val_names = get_past_key_names() + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=2, model_split_map=MODEL_SPLIT_MAP + ) + past_key_val_names = get_past_key_names( + start=layers_start, + end=layers_end, + num_of_past_key_heads=NUM_KEY_VAL_HEADS, + suffix="_in", + ) for past_key_val in past_key_val_names: if "key" in past_key_val: input_spec[past_key_val] = ( @@ -976,6 +1061,17 @@ def get_input_spec( ) return input_spec + @staticmethod + def get_output_names(): + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=2, model_split_map=MODEL_SPLIT_MAP + ) + return Llama_QuantizedMixin.get_output_names( + start=layers_start, + end=layers_end, + past_key_val_heads=NUM_KEY_VAL_HEADS, + ) + @staticmethod def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): data = load_input_cached_data( @@ -1005,13 +1101,18 @@ def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): del model data = { - "input_ids": output_tg[0].detach(), + _get_intermediate_logit_name(split_part=2): output_tg[0].detach(), "attention_mask": inputs["attention_mask"], "position_ids_cos": inputs["position_ids_cos"], "position_ids_sin": inputs["position_ids_sin"], } - key_val = get_past_keyval_with_shift(output[1:], NUM_KEY_VAL_HEADS) + layers_start, _ = get_hidden_layer_range_from_split( + split_part=2, model_split_map=MODEL_SPLIT_MAP + ) + key_val = get_past_keyval_with_shift( + output[1:], layers_start, NUM_KEY_VAL_HEADS, new_key_suffix="_in" + ) for key, val in key_val.items(): data[key] = val @@ -1086,14 +1187,25 @@ def get_input_spec( # the model input specification upon submitting a compile job. input_spec = { - "input_ids": ((1, 1, ATTENTION_HIDDEN_DIM), "float32"), + _get_intermediate_logit_name(split_part=3): ( + (1, 1, ATTENTION_HIDDEN_DIM), + "float32", + ), "attention_mask": ((1, 1, 1, input_seq_length), "float32"), "position_ids_cos": ((1, 1, 1, POS_EMBED_DIM), "float32"), "position_ids_sin": ((1, 1, 1, POS_EMBED_DIM), "float32"), } # Collect past_key_values and drop output names - past_key_val_names = get_past_key_names() + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=3, model_split_map=MODEL_SPLIT_MAP + ) + past_key_val_names = get_past_key_names( + start=layers_start, + end=layers_end, + num_of_past_key_heads=NUM_KEY_VAL_HEADS, + suffix="_in", + ) for past_key_val in past_key_val_names: if "key" in past_key_val: input_spec[past_key_val] = ( @@ -1107,6 +1219,17 @@ def get_input_spec( ) return input_spec + @staticmethod + def get_output_names(): + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=3, model_split_map=MODEL_SPLIT_MAP + ) + return Llama_QuantizedMixin.get_output_names( + start=layers_start, + end=layers_end, + past_key_val_heads=NUM_KEY_VAL_HEADS, + ) + @staticmethod def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): data = load_input_cached_data( @@ -1136,13 +1259,18 @@ def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): del model data = { - "input_ids": output_tg[0].detach(), + _get_intermediate_logit_name(split_part=3): output_tg[0].detach(), "attention_mask": inputs["attention_mask"], "position_ids_cos": inputs["position_ids_cos"], "position_ids_sin": inputs["position_ids_sin"], } - key_val = get_past_keyval_with_shift(output[1:], NUM_KEY_VAL_HEADS) + layers_start, _ = get_hidden_layer_range_from_split( + split_part=3, model_split_map=MODEL_SPLIT_MAP + ) + key_val = get_past_keyval_with_shift( + output[1:], layers_start, NUM_KEY_VAL_HEADS, new_key_suffix="_in" + ) for key, val in key_val.items(): data[key] = val @@ -1217,14 +1345,25 @@ def get_input_spec( # the model input specification upon submitting a compile job. input_spec = { - "input_ids": ((1, 1, ATTENTION_HIDDEN_DIM), "float32"), + _get_intermediate_logit_name(split_part=4): ( + (1, 1, ATTENTION_HIDDEN_DIM), + "float32", + ), "attention_mask": ((1, 1, 1, input_seq_length), "float32"), "position_ids_cos": ((1, 1, 1, POS_EMBED_DIM), "float32"), "position_ids_sin": ((1, 1, 1, POS_EMBED_DIM), "float32"), } # Collect past_key_values and drop output names - past_key_val_names = get_past_key_names() + layers_start, layers_end = get_hidden_layer_range_from_split( + split_part=4, model_split_map=MODEL_SPLIT_MAP + ) + past_key_val_names = get_past_key_names( + start=layers_start, + end=layers_end, + num_of_past_key_heads=NUM_KEY_VAL_HEADS, + suffix="_in", + ) for past_key_val in past_key_val_names: if "key" in past_key_val: input_spec[past_key_val] = ( @@ -1279,13 +1418,18 @@ def get_model_data(input_seq_len: int = DEFAULT_INPUT_SEQ_LEN): del model data = { - "input_ids": output_tg[0].detach(), + _get_intermediate_logit_name(split_part=4): output_tg[0].detach(), "attention_mask": inputs["attention_mask"], "position_ids_cos": inputs["position_ids_cos"], "position_ids_sin": inputs["position_ids_sin"], } - key_val = get_past_keyval_with_shift(output[1:], NUM_KEY_VAL_HEADS) + layers_start, _ = get_hidden_layer_range_from_split( + split_part=4, model_split_map=MODEL_SPLIT_MAP + ) + key_val = get_past_keyval_with_shift( + output[1:], layers_start, NUM_KEY_VAL_HEADS, new_key_suffix="_in" + ) for key, val in key_val.items(): data[key] = val diff --git a/qai_hub_models/models/mediapipe_face/demo.py b/qai_hub_models/models/mediapipe_face/demo.py index 53310133..8dd031e2 100644 --- a/qai_hub_models/models/mediapipe_face/demo.py +++ b/qai_hub_models/models/mediapipe_face/demo.py @@ -3,6 +3,7 @@ # SPDX-License-Identifier: BSD-3-Clause # --------------------------------------------------------------------- import argparse +from typing import Type import numpy as np from PIL import Image @@ -25,7 +26,7 @@ # Run Mediapipe Face landmark detection end-to-end on a sample image or camera stream. # The demo will display output with the predicted landmarks & bounding boxes drawn. -def main(is_test: bool = False): +def mediapipe_face_demo(model_cls: Type[MediaPipeFace], is_test: bool = False): # Demo parameters parser = argparse.ArgumentParser() parser.add_argument( @@ -65,7 +66,7 @@ def main(is_test: bool = False): # Load app app = MediaPipeFaceApp( - MediaPipeFace.from_pretrained(), + model_cls.from_pretrained(), args.score_threshold, args.iou_threshold, ) @@ -87,5 +88,9 @@ def frame_processor(frame: np.ndarray) -> np.ndarray: ) +def main(is_test: bool = False): + return mediapipe_face_demo(MediaPipeFace, is_test) + + if __name__ == "__main__": main() diff --git a/qai_hub_models/models/mediapipe_face/export.py b/qai_hub_models/models/mediapipe_face/export.py index ec0361e4..b193a770 100644 --- a/qai_hub_models/models/mediapipe_face/export.py +++ b/qai_hub_models/models/mediapipe_face/export.py @@ -128,8 +128,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/mediapipe_face/model.py b/qai_hub_models/models/mediapipe_face/model.py index 89406fc8..0b4967a3 100644 --- a/qai_hub_models/models/mediapipe_face/model.py +++ b/qai_hub_models/models/mediapipe_face/model.py @@ -9,6 +9,12 @@ import torch from qai_hub_models.models._shared.mediapipe.utils import MediaPipePyTorchAsRoot +from qai_hub_models.models.common import SampleInputsType +from qai_hub_models.utils.asset_loaders import ( + CachedWebModelAsset, + find_replace_in_repo, + load_numpy, +) from qai_hub_models.utils.base_model import BaseModel, CollectionModel from qai_hub_models.utils.input_spec import InputSpec @@ -164,6 +170,12 @@ ROTATION_VECTOR_OFFSET_RADS = ( 0 # Offset required when computing rotation of the detected face. ) +FACE_DETECTOR_SAMPLE_INPUTS_ADDRESS = CachedWebModelAsset.from_asset_store( + MODEL_ID, MODEL_ASSET_VERSION, "face_detector_inputs.npy" +) +LANDMARK_DETECTOR_SAMPLE_INPUTS_ADDRESS = CachedWebModelAsset.from_asset_store( + MODEL_ID, MODEL_ASSET_VERSION, "landmark_detector_inputs.npy" +) class MediaPipeFace(CollectionModel): @@ -203,7 +215,12 @@ def from_pretrained( .blazeface_landmark.BlazeFaceLandmark, ] """ - with MediaPipePyTorchAsRoot(): + with MediaPipePyTorchAsRoot() as repo_path: + # This conditional is unlikely to be hit, and breaks torch fx graph conversion + find_replace_in_repo( + repo_path, "blazeface_landmark.py", "if x.shape[0] == 0:", "if False:" + ) + from blazeface import BlazeFace from blazeface_landmark import BlazeFaceLandmark @@ -229,7 +246,7 @@ def __init__( self.detector = detector self.anchors = anchors - def forward(self, image: torch.Tensor): + def forward(self, image): return self.detector(image) @classmethod @@ -237,7 +254,7 @@ def from_pretrained( cls, detector_weights: str = "blazefaceback.pth", detector_anchors: str = "anchors_face_back.npy", - ): + ) -> FaceDetector: with MediaPipePyTorchAsRoot(): from blazeface import BlazeFace @@ -262,6 +279,11 @@ def get_output_names() -> List[str]: def get_channel_last_inputs() -> List[str]: return ["image"] + def _sample_inputs_impl( + self, input_spec: InputSpec | None = None + ) -> SampleInputsType: + return {"image": [load_numpy(FACE_DETECTOR_SAMPLE_INPUTS_ADDRESS)]} + class FaceLandmarkDetector(BaseModel): def __init__( @@ -271,12 +293,17 @@ def __init__( super().__init__() self.detector = detector - def forward(self, image: torch.Tensor): + def forward(self, image): return self.detector(image) @classmethod def from_pretrained(cls, landmark_detector_weights: str = "blazeface_landmark.pth"): - with MediaPipePyTorchAsRoot(): + with MediaPipePyTorchAsRoot() as repo_path: + # This conditional is unlikely to be hit, and breaks torch fx graph conversion + find_replace_in_repo( + repo_path, "blazeface_landmark.py", "if x.shape[0] == 0:", "if False:" + ) + from blazeface_landmark import BlazeFaceLandmark face_regressor = BlazeFaceLandmark() @@ -298,3 +325,8 @@ def get_output_names() -> List[str]: @staticmethod def get_channel_last_inputs() -> List[str]: return ["image"] + + def _sample_inputs_impl( + self, input_spec: InputSpec | None = None + ) -> SampleInputsType: + return {"image": [load_numpy(LANDMARK_DETECTOR_SAMPLE_INPUTS_ADDRESS)]} diff --git a/qai_hub_models/models/mediapipe_face/perf.yaml b/qai_hub_models/models/mediapipe_face/perf.yaml index 24ae3f78..5a96aebd 100644 --- a/qai_hub_models/models/mediapipe_face/perf.yaml +++ b/qai_hub_models/models/mediapipe_face/perf.yaml @@ -45,11 +45,11 @@ models: - name: MediaPipeFaceDetector performance_metrics: - torchscript_onnx_tflite: - inference_time: 577.0 - throughput: 1733.102253032929 + inference_time: 547.0 + throughput: 1828.1535648994516 estimated_peak_memory_range: - min: 12288 - max: 1513296 + min: 28672 + max: 1458784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 111 - job_id: jz5wyzm3g + job_id: jz5wrmo3p job_status: Passed torchscript_onnx_qnn: - inference_time: 638.0 - throughput: 1567.398119122257 + inference_time: 622.0 + throughput: 1607.717041800643 estimated_peak_memory_range: - min: 12288 - max: 5302104 + min: 16384 + max: 6313472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,22 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jogk60rw5 - job_status: Passed - torchscript_onnx: - inference_time: 1070.0 - throughput: 934.5794392523364 - estimated_peak_memory_range: - min: 274432 - max: 2036176 - primary_compute_unit: NPU - precision: fp16 - layer_info: - layers_on_npu: 147 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 147 - job_id: jnp1o1375 + job_id: jogkkrzwg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +81,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:23:26Z' + timestamp: '2024-08-27T00:37:10Z' - torchscript_onnx_tflite: - inference_time: 411.0 - throughput: 2433.0900243309 + inference_time: 405.0 + throughput: 2469.135802469136 estimated_peak_memory_range: min: 12288 - max: 32788480 + max: 33526544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +95,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 111 - job_id: jnp1o1q85 + job_id: jnp1mq08p job_status: Passed torchscript_onnx_qnn: - inference_time: 461.0 - throughput: 2169.1973969631235 + inference_time: 459.0 + throughput: 2178.649237472767 estimated_peak_memory_range: - min: 802816 - max: 15976416 + min: 0 + max: 16371040 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: j1glw8ejp + job_id: j1glqenjp job_status: Passed torchscript_onnx: - inference_time: 756.0 - throughput: 1322.7513227513227 + inference_time: 754.0 + throughput: 1326.2599469496022 estimated_peak_memory_range: min: 0 - max: 37318800 + max: 38067168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +125,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jz57on69g + job_id: jegnwrykg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +134,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:23:28Z' + timestamp: '2024-08-27T00:37:26Z' - torchscript_onnx_tflite: - inference_time: 584.0 - throughput: 1712.3287671232877 + inference_time: 539.0 + throughput: 1855.287569573284 estimated_peak_memory_range: min: 12288 - max: 1352792 + max: 1825512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +148,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 111 - job_id: jz57onvvg + job_id: jz578vzvp job_status: Passed torchscript_onnx_qnn: - inference_time: 594.0 - throughput: 1683.5016835016836 + inference_time: 599.0 + throughput: 1669.449081803005 estimated_peak_memory_range: min: 811008 - max: 2005592 + max: 2047792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +163,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: j1pv2m4kg + job_id: j1pvnzokg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +172,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:23:15Z' + timestamp: '2024-08-27T00:37:15Z' - torchscript_onnx_tflite: - inference_time: 832.0 - throughput: 1201.923076923077 + inference_time: 755.0 + throughput: 1324.5033112582782 estimated_peak_memory_range: - min: 16384 - max: 31463840 + min: 12288 + max: 31762064 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +186,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 111 - job_id: j0px0re3p + job_id: j0pxzev35 job_status: Passed torchscript_onnx_qnn: - inference_time: 827.0 - throughput: 1209.1898428053205 + inference_time: 825.0 + throughput: 1212.121212121212 estimated_peak_memory_range: min: 802816 - max: 16453984 + max: 15195888 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +201,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jz5wyzxmg + job_id: jz578vwvp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +210,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:23:24Z' + timestamp: '2024-08-27T00:37:23Z' - torchscript_onnx_tflite: - inference_time: 584.0 - throughput: 1712.3287671232877 + inference_time: 548.0 + throughput: 1824.8175182481752 estimated_peak_memory_range: - min: 24576 - max: 1615192 + min: 12288 + max: 59011704 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +224,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 111 - job_id: jegn1qrkp + job_id: jegnwr2kg job_status: Passed torchscript_onnx_qnn: - inference_time: 598.0 - throughput: 1672.2408026755852 + inference_time: 613.0 + throughput: 1631.3213703099511 estimated_peak_memory_range: min: 827392 - max: 2159792 + max: 1996992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +239,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jlpe6x2og + job_id: jlpen41op job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +248,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:23:17Z' + timestamp: '2024-08-27T00:37:17Z' - torchscript_onnx_tflite: - inference_time: 590.0 - throughput: 1694.915254237288 + inference_time: 544.0 + throughput: 1838.235294117647 estimated_peak_memory_range: - min: 20480 - max: 1358592 + min: 28672 + max: 1328256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +262,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 111 - job_id: jep2od3rg + job_id: jep2z38rp job_status: Passed torchscript_onnx_qnn: - inference_time: 606.0 - throughput: 1650.1650165016501 + inference_time: 613.0 + throughput: 1631.3213703099511 estimated_peak_memory_range: - min: 839680 - max: 2119440 + min: 819200 + max: 2032240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +277,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jz5wyzx3g + job_id: jz5wrmv3p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +286,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:23:19Z' + timestamp: '2024-08-27T00:37:19Z' - torchscript_onnx_tflite: - inference_time: 576.0 - throughput: 1736.111111111111 + inference_time: 547.0 + throughput: 1828.1535648994516 estimated_peak_memory_range: - min: 12288 - max: 1309040 + min: 36864 + max: 1995176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +300,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 111 - job_id: j2p0o9e9p + job_id: j2p0xey9p job_status: Passed torchscript_onnx_qnn: - inference_time: 608.0 - throughput: 1644.7368421052631 + inference_time: 602.0 + throughput: 1661.1295681063123 estimated_peak_memory_range: - min: 823296 - max: 2654992 + min: 815104 + max: 2471992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +315,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jnp1o1385 + job_id: jnp1mql8p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +324,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:23:22Z' + timestamp: '2024-08-27T00:37:21Z' - torchscript_onnx_qnn: - inference_time: 763.0 - throughput: 1310.615989515072 + inference_time: 766.0 + throughput: 1305.4830287206266 estimated_peak_memory_range: min: 786432 max: 786432 @@ -353,14 +338,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: j1p3o7q3p + job_id: j1p3rqk3p job_status: Passed torchscript_onnx: - inference_time: 1052.0 - throughput: 950.5703422053232 + inference_time: 1091.0 + throughput: 916.5902841429881 estimated_peak_memory_range: - min: 1978368 - max: 1978368 + min: 2015232 + max: 2015232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +353,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: j0px0rmlp + job_id: jep2z36rp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,15 +362,15 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:23:29Z' + timestamp: '2024-08-27T00:37:27Z' - name: MediaPipeFaceLandmarkDetector performance_metrics: - torchscript_onnx_tflite: - inference_time: 207.0 - throughput: 4830.917874396136 + inference_time: 189.0 + throughput: 5291.005291005291 estimated_peak_memory_range: - min: 12288 - max: 13882888 + min: 16384 + max: 1369760 primary_compute_unit: NPU precision: fp16 layer_info: @@ -393,14 +378,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 100 - job_id: jmg9o29wg + job_id: jmg9q9vwp job_status: Passed torchscript_onnx_qnn: - inference_time: 285.0 - throughput: 3508.7719298245615 + inference_time: 286.0 + throughput: 3496.5034965034965 estimated_peak_memory_range: - min: 2125824 - max: 10209168 + min: 471040 + max: 3620192 primary_compute_unit: NPU precision: fp16 layer_info: @@ -408,14 +393,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jn5q419ng + job_id: jn5qd98ng job_status: Passed torchscript_onnx: - inference_time: 490.0 - throughput: 2040.8163265306123 + inference_time: 508.0 + throughput: 1968.5039370078741 estimated_peak_memory_range: min: 12288 - max: 76839048 + max: 2783856 primary_compute_unit: NPU precision: fp16 layer_info: @@ -423,7 +408,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: jvgd640zp + job_id: jo5mlv2dg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -432,13 +417,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:23:26Z' + timestamp: '2024-08-27T00:37:25Z' - torchscript_onnx_tflite: - inference_time: 158.0 - throughput: 6329.113924050633 + inference_time: 146.0 + throughput: 6849.315068493151 estimated_peak_memory_range: - min: 20480 - max: 28835072 + min: 16384 + max: 29368512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -446,14 +431,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 100 - job_id: jvgd647rp + job_id: jvgdm7wrg job_status: Passed torchscript_onnx_qnn: - inference_time: 210.0 - throughput: 4761.9047619047615 + inference_time: 216.0 + throughput: 4629.62962962963 estimated_peak_memory_range: min: 458752 - max: 12670896 + max: 12650880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -461,14 +446,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jw56omq65 + job_id: jw560q665 job_status: Passed torchscript_onnx: - inference_time: 382.0 - throughput: 2617.801047120419 + inference_time: 352.0 + throughput: 2840.909090909091 estimated_peak_memory_range: - min: 0 - max: 31570112 + min: 278528 + max: 32345248 primary_compute_unit: NPU precision: fp16 layer_info: @@ -476,7 +461,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: jqp4e481g + job_id: jopr71q0g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -485,13 +470,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:23:28Z' + timestamp: '2024-08-27T00:37:26Z' - torchscript_onnx_tflite: - inference_time: 202.0 - throughput: 4950.495049504951 + inference_time: 191.0 + throughput: 5235.602094240838 estimated_peak_memory_range: - min: 12288 - max: 11811984 + min: 16384 + max: 4248952 primary_compute_unit: NPU precision: fp16 layer_info: @@ -499,14 +484,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 100 - job_id: jqp4e4j8g + job_id: jqp42jq8g job_status: Passed torchscript_onnx_qnn: - inference_time: 278.0 - throughput: 3597.122302158273 + inference_time: 276.0 + throughput: 3623.1884057971015 estimated_peak_memory_range: - min: 475136 - max: 1661032 + min: 466944 + max: 1620400 primary_compute_unit: NPU precision: fp16 layer_info: @@ -514,7 +499,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: j7gj3y1vp + job_id: j7gj8kmv5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -523,13 +508,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:23:15Z' + timestamp: '2024-08-27T00:37:16Z' - torchscript_onnx_tflite: - inference_time: 324.0 - throughput: 3086.41975308642 + inference_time: 272.0 + throughput: 3676.470588235294 estimated_peak_memory_range: - min: 16384 - max: 29931536 + min: 0 + max: 29153568 primary_compute_unit: NPU precision: fp16 layer_info: @@ -537,14 +522,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 100 - job_id: jo5m9kvdg + job_id: jo5mlvrdg job_status: Passed torchscript_onnx_qnn: - inference_time: 369.0 - throughput: 2710.027100271003 + inference_time: 359.0 + throughput: 2785.515320334262 estimated_peak_memory_range: - min: 458752 - max: 14424512 + min: 462848 + max: 13302864 primary_compute_unit: NPU precision: fp16 layer_info: @@ -552,7 +537,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jmg9o288g + job_id: jqp42jo8g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -561,13 +546,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:23:24Z' + timestamp: '2024-08-27T00:37:23Z' - torchscript_onnx_tflite: - inference_time: 213.0 - throughput: 4694.835680751174 + inference_time: 189.0 + throughput: 5291.005291005291 estimated_peak_memory_range: min: 24576 - max: 1507472 + max: 1909936 primary_compute_unit: NPU precision: fp16 layer_info: @@ -575,14 +560,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 100 - job_id: joprxd10p + job_id: jopr71k0g job_status: Passed torchscript_onnx_qnn: inference_time: 279.0 throughput: 3584.2293906810037 estimated_peak_memory_range: - min: 475136 - max: 2113168 + min: 487424 + max: 2080168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -590,7 +575,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jygzzywog + job_id: jygz0v9o5 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -599,13 +584,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:23:18Z' + timestamp: '2024-08-27T00:37:17Z' - torchscript_onnx_tflite: - inference_time: 213.0 - throughput: 4694.835680751174 + inference_time: 186.0 + throughput: 5376.344086021505 estimated_peak_memory_range: - min: 16384 - max: 1458072 + min: 24576 + max: 1596368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -613,14 +598,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 100 - job_id: jqpy82v8g + job_id: jqpyyve8p job_status: Passed torchscript_onnx_qnn: - inference_time: 282.0 - throughput: 3546.099290780142 + inference_time: 274.0 + throughput: 3649.6350364963505 estimated_peak_memory_range: - min: 475136 - max: 1765040 + min: 471040 + max: 1877832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -628,7 +613,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jmg9o28wg + job_id: jmg9q91wp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -637,13 +622,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:23:20Z' + timestamp: '2024-08-27T00:37:19Z' - torchscript_onnx_tflite: - inference_time: 197.0 - throughput: 5076.1421319796955 + inference_time: 193.0 + throughput: 5181.347150259067 estimated_peak_memory_range: - min: 77824 - max: 1371112 + min: 20480 + max: 1473424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -651,14 +636,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 100 - job_id: j1p8jrwk5 + job_id: j1p8kwokp job_status: Passed torchscript_onnx_qnn: - inference_time: 280.0 - throughput: 3571.4285714285716 + inference_time: 277.0 + throughput: 3610.1083032490974 estimated_peak_memory_range: - min: 471040 - max: 1685560 + min: 466944 + max: 1776712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -666,7 +651,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jvgd640rp + job_id: jvgdm79rg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -675,13 +660,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:23:22Z' + timestamp: '2024-08-27T00:37:21Z' - torchscript_onnx_qnn: - inference_time: 368.0 - throughput: 2717.391304347826 + inference_time: 455.0 + throughput: 2197.802197802198 estimated_peak_memory_range: - min: 442368 - max: 442368 + min: 1462272 + max: 1462272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -689,14 +674,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jwgodweq5 + job_id: jwgo9eyqg job_status: Passed torchscript_onnx: - inference_time: 531.0 - throughput: 1883.2391713747645 + inference_time: 513.0 + throughput: 1949.317738791423 estimated_peak_memory_range: - min: 3055616 - max: 3055616 + min: 4255744 + max: 4255744 primary_compute_unit: NPU precision: fp16 layer_info: @@ -704,7 +689,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: jo5m9k49g + job_id: jqpyyvw8p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -713,4 +698,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:23:30Z' + timestamp: '2024-08-27T00:37:28Z' diff --git a/qai_hub_models/models/mediapipe_face_quantized/README.md b/qai_hub_models/models/mediapipe_face_quantized/README.md new file mode 100644 index 00000000..72472470 --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/README.md @@ -0,0 +1,55 @@ +[![Qualcomm® AI Hub Models](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/quic-logo.jpg)](../../README.md) + + +# [MediaPipe-Face-Detection-Quantized: Detect faces and locate facial features in real-time video and image streams](https://aihub.qualcomm.com/models/mediapipe_face_quantized) + +Designed for sub-millisecond processing, this model predicts bounding boxes and pose skeletons (left eye, right eye, nose tip, mouth, left eye tragion, and right eye tragion) of faces in an image. + +This is based on the implementation of MediaPipe-Face-Detection-Quantized found +[here](https://github.com/zmurez/MediaPipePyTorch/). This repository contains scripts for optimized on-device +export suitable to run on Qualcomm® devices. More details on model performance +accross various devices, can be found [here](https://aihub.qualcomm.com/models/mediapipe_face_quantized). + +[Sign up](https://myaccount.qualcomm.com/signup) to start using Qualcomm AI Hub and run these models on a hosted Qualcomm® device. + + + + +## Example & Usage + + +Once installed, run the following simple CLI demo: + +```bash +python -m qai_hub_models.models.mediapipe_face_quantized.demo +``` +More details on the CLI tool can be found with the `--help` option. See +[demo.py](demo.py) for sample usage of the model including pre/post processing +scripts. Please refer to our [general instructions on using +models](../../../#getting-started) for more usage instructions. + +## Export for on-device deployment + +This repository contains export scripts that produce a model optimized for +on-device deployment. This can be run as follows: + +```bash +python -m qai_hub_models.models.mediapipe_face_quantized.export +``` +Additional options are documented with the `--help` option. Note that the above +script requires access to Deployment instructions for Qualcomm® AI Hub. + +## License +- The license for the original implementation of MediaPipe-Face-Detection-Quantized can be found + [here](https://github.com/zmurez/MediaPipePyTorch/blob/master/LICENSE). +- The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf) + +## References +* [BlazeFace: Sub-millisecond Neural Face Detection on Mobile GPUs](https://arxiv.org/abs/1907.05047) +* [Source Model Implementation](https://github.com/zmurez/MediaPipePyTorch/) + +## Community +* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI. +* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com). + + diff --git a/qai_hub_models/models/mediapipe_face_quantized/__init__.py b/qai_hub_models/models/mediapipe_face_quantized/__init__.py new file mode 100644 index 00000000..03a92941 --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/__init__.py @@ -0,0 +1,10 @@ +# --------------------------------------------------------------------- +# Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# --------------------------------------------------------------------- +from qai_hub_models.models.mediapipe_face.app import ( # noqa: F401 + MediaPipeFaceApp as App, +) + +from .model import MODEL_ID # noqa: F401 +from .model import MediaPipeFaceQuantizable as Model # noqa: F401 diff --git a/qai_hub_models/models/mediapipe_face_quantized/conftest.py b/qai_hub_models/models/mediapipe_face_quantized/conftest.py new file mode 100644 index 00000000..02dd12e6 --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/conftest.py @@ -0,0 +1,39 @@ +# --------------------------------------------------------------------- +# Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# --------------------------------------------------------------------- +# THIS FILE WAS AUTO-GENERATED. DO NOT EDIT MANUALLY. + +import inspect + +import pytest + +from qai_hub_models.models.mediapipe_face_quantized import Model +from qai_hub_models.utils.testing import skip_clone_repo_check + + +# Instantiate the model only once for all tests. +# Mock from_pretrained to always return the initialized model. +# This speeds up tests and limits memory leaks. +@pytest.fixture(scope="module", autouse=True) +def cached_from_pretrained(): + with pytest.MonkeyPatch.context() as mp: + pretrained_cache = {} + from_pretrained = Model.from_pretrained + sig = inspect.signature(from_pretrained) + + @skip_clone_repo_check + def _cached_from_pretrained(*args, **kwargs): + cache_key = str(args) + str(kwargs) + model = pretrained_cache.get(cache_key, None) + if model: + return model + else: + model = from_pretrained(*args, **kwargs) + pretrained_cache[cache_key] = model + return model + + _cached_from_pretrained.__signature__ = sig + + mp.setattr(Model, "from_pretrained", _cached_from_pretrained) + yield mp diff --git a/qai_hub_models/models/mediapipe_face_quantized/demo.py b/qai_hub_models/models/mediapipe_face_quantized/demo.py new file mode 100644 index 00000000..150f6e4b --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/demo.py @@ -0,0 +1,16 @@ +# --------------------------------------------------------------------- +# Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# --------------------------------------------------------------------- +from qai_hub_models.models.mediapipe_face.demo import mediapipe_face_demo +from qai_hub_models.models.mediapipe_face_quantized.model import ( + MediaPipeFaceQuantizable, +) + + +def main(is_test: bool = False): + return mediapipe_face_demo(MediaPipeFaceQuantizable, is_test) + + +if __name__ == "__main__": + main() diff --git a/qai_hub_models/models/mediapipe_face_quantized/export.py b/qai_hub_models/models/mediapipe_face_quantized/export.py new file mode 100644 index 00000000..5a0d1ad2 --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/export.py @@ -0,0 +1,264 @@ +# --------------------------------------------------------------------- +# Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# --------------------------------------------------------------------- +# THIS FILE WAS AUTO-GENERATED. DO NOT EDIT MANUALLY. + + +from __future__ import annotations + +import os +import warnings +from pathlib import Path +from typing import Any, Dict, List, Mapping, Optional, Tuple, cast + +import qai_hub as hub + +from qai_hub_models.models.mediapipe_face_quantized import Model +from qai_hub_models.utils.args import export_parser, get_model_kwargs +from qai_hub_models.utils.base_model import BaseModel, TargetRuntime +from qai_hub_models.utils.compare import torch_inference +from qai_hub_models.utils.printing import ( + print_inference_metrics, + print_profile_metrics_from_job, +) +from qai_hub_models.utils.qai_hub_helpers import ( + can_access_qualcomm_ai_hub, + export_without_hub_access, +) + +ALL_COMPONENTS = ["MediaPipeFaceDetector", "MediaPipeFaceLandmarkDetector"] + + +def export_model( + device: str = "Samsung Galaxy S23 (Family)", + chipset: Optional[str] = None, + components: Optional[List[str]] = None, + skip_profiling: bool = False, + skip_inferencing: bool = False, + skip_downloading: bool = False, + skip_summary: bool = False, + output_dir: Optional[str] = None, + target_runtime: TargetRuntime = TargetRuntime.TFLITE, + compile_options: str = "", + profile_options: str = "", + **additional_model_kwargs, +) -> Mapping[ + str, Tuple[hub.CompileJob, Optional[hub.ProfileJob], Optional[hub.InferenceJob]] +] | List[str]: + """ + This function accomplishes 6 main tasks: + + 1. Instantiates a PyTorch model and converts it to a traced TorchScript format. + 2. Compiles the model to an asset that can be run on device. + 3. Profiles the model performance on real devices. + 4. Inferences the model on sample inputs. + 5. Downloads the model asset to the local directory. + 6. Summarizes the results from profiling and inference. + + Each of the last four steps can be optionally skipped using the input options. + + Parameters: + device: Device for which to export the model. + Full list of available devices can be found by running `hub.get_devices()`. + Defaults to DEFAULT_DEVICE if not specified. + chipset: If set, will choose a random device with this chipset. + Overrides the `device` argument. + components: List of sub-components of the model that will be exported. + Each component is compiled and profiled separately. + Defaults to ALL_COMPONENTS if not specified. + skip_profiling: If set, skips profiling of compiled model on real devices. + skip_inferencing: If set, skips computing on-device outputs from sample data. + skip_downloading: If set, skips downloading of compiled model. + skip_summary: If set, skips waiting for and summarizing results + from profiling and inference. + output_dir: Directory to store generated assets (e.g. compiled model). + Defaults to `/build/`. + target_runtime: Which on-device runtime to target. Default is TFLite. + compile_options: Additional options to pass when submitting the compile job. + profile_options: Additional options to pass when submitting the profile job. + **additional_model_kwargs: Additional optional kwargs used to customize + `model_cls.from_pretrained` + + Returns: + A Mapping from component_name to a 3-tuple of: + * A CompileJob object containing metadata about the compile job submitted to hub. + * A ProfileJob containing metadata about the profile job (None if profiling skipped). + * An InferenceJob containing metadata about the inference job (None if inferencing skipped). + """ + model_name = "mediapipe_face_quantized" + output_path = Path(output_dir or Path.cwd() / "build" / model_name) + if chipset: + hub_device = hub.Device(attributes=f"chipset:{chipset}") + else: + hub_device = hub.Device(name=device) + component_arg = components + components = components or ALL_COMPONENTS + for component_name in components: + if component_name not in ALL_COMPONENTS: + raise ValueError(f"Invalid component {component_name}.") + if not can_access_qualcomm_ai_hub(): + return export_without_hub_access( + "mediapipe_face_quantized", + "MediaPipe-Face-Detection-Quantized", + device, + skip_profiling, + skip_inferencing, + skip_downloading, + skip_summary, + output_path, + target_runtime, + compile_options, + profile_options, + component_arg, + ) + + # On-device perf improves with I/O in channel_last format except when using ONNX. + use_channel_last_format = target_runtime != TargetRuntime.ONNX + + # 1. Initialize PyTorch model + model = Model.from_pretrained(**get_model_kwargs(Model, additional_model_kwargs)) + components_dict: Dict[str, BaseModel] = {} + if "MediaPipeFaceDetector" in components: + components_dict["MediaPipeFaceDetector"] = model.face_detector # type: ignore + if "MediaPipeFaceLandmarkDetector" in components: + components_dict["MediaPipeFaceLandmarkDetector"] = model.face_landmark_detector # type: ignore + + compile_jobs: Dict[str, hub.client.CompileJob] = {} + for component_name, component in components_dict.items(): + input_spec = component.get_input_spec() + + # Trace the model + source_model = component.convert_to_hub_source_model( + target_runtime, output_path, input_spec + ) + if target_runtime == TargetRuntime.TFLITE: + quant_calibration_data = None + else: + quant_calibration_data = component.get_calibration_data( + target_runtime, input_spec + ) + + # 2. Compile the models to an on-device asset + model_compile_options = component.get_hub_compile_options( + target_runtime, compile_options, hub_device + ) + print(f"Optimizing model {component_name} to run on-device") + submitted_compile_job = hub.submit_compile_job( + model=source_model, + input_specs=input_spec, + device=hub_device, + name=f"{model_name}_{component_name}", + calibration_data=quant_calibration_data, + options=model_compile_options, + ) + compile_jobs[component_name] = cast( + hub.client.CompileJob, submitted_compile_job + ) + + # 3. Profile the model assets on real devices + profile_jobs: Dict[str, hub.client.ProfileJob] = {} + if not skip_profiling: + for component_name in components: + profile_options_all = components_dict[ + component_name + ].get_hub_profile_options(target_runtime, profile_options) + print(f"Profiling model {component_name} on a hosted device.") + submitted_profile_job = hub.submit_profile_job( + model=compile_jobs[component_name].get_target_model(), + device=hub_device, + name=f"{model_name}_{component_name}", + options=profile_options_all, + ) + profile_jobs[component_name] = cast( + hub.client.ProfileJob, submitted_profile_job + ) + + # 4. Run inference on-device with sample inputs + inference_jobs: Dict[str, hub.client.InferenceJob] = {} + if not skip_inferencing: + for component_name in components: + print( + f"Running inference for {component_name} on a hosted device with example inputs." + ) + profile_options_all = components_dict[ + component_name + ].get_hub_profile_options(target_runtime, profile_options) + sample_inputs = components_dict[component_name].sample_inputs( + use_channel_last_format=use_channel_last_format + ) + submitted_inference_job = hub.submit_inference_job( + model=compile_jobs[component_name].get_target_model(), + inputs=sample_inputs, + device=hub_device, + name=f"{model_name}_{component_name}", + options=profile_options_all, + ) + inference_jobs[component_name] = cast( + hub.client.InferenceJob, submitted_inference_job + ) + + # 5. Download the model assets to a local file + if not skip_downloading: + if target_runtime == TargetRuntime.QNN: + target_runtime_extension = "so" + elif target_runtime == TargetRuntime.TFLITE: + target_runtime_extension = "tflite" + elif target_runtime in {TargetRuntime.ONNX, TargetRuntime.PRECOMPILED_QNN_ONNX}: + target_runtime_extension = "onnx" + + os.makedirs(output_path, exist_ok=True) + for component_name, compile_job in compile_jobs.items(): + target_model: hub.Model = compile_job.get_target_model() # type: ignore + target_model.download( + str( + output_path + / f"{model_name}_{component_name}.{target_runtime_extension}" + ) + ) + + # 6. Summarize the results from profiling and inference + if not skip_summary and not skip_profiling: + for component_name in components: + profile_job = profile_jobs[component_name] + assert profile_job is not None and profile_job.wait().success + profile_data: Dict[str, Any] = profile_job.download_profile() # type: ignore + print_profile_metrics_from_job(profile_job, profile_data) + + if not skip_summary and not skip_inferencing: + for component_name, component in components_dict.items(): + inference_job = inference_jobs[component_name] + sample_inputs = component.sample_inputs(use_channel_last_format=False) + torch_out = torch_inference( + component, + sample_inputs, + return_channel_last_output=use_channel_last_format, + ) + assert inference_job is not None and inference_job.wait().success + inference_result: hub.client.DatasetEntries = inference_job.download_output_data() # type: ignore + + print_inference_metrics( + inference_job, inference_result, torch_out, component.get_output_names() + ) + + return { + component_name: ( + compile_jobs[component_name], + profile_jobs.get(component_name, None), + inference_jobs.get(component_name, None), + ) + for component_name in components + } + + +def main(): + warnings.filterwarnings("ignore") + parser = export_parser( + model_cls=Model, components=ALL_COMPONENTS, supports_onnx=False + ) + args = parser.parse_args() + export_model(**vars(args)) + + +if __name__ == "__main__": + main() diff --git a/qai_hub_models/models/mediapipe_face_quantized/info.yaml b/qai_hub_models/models/mediapipe_face_quantized/info.yaml new file mode 100644 index 00000000..806d2c92 --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/info.yaml @@ -0,0 +1,44 @@ +name: MediaPipe-Face-Detection-Quantized +# id must match with the model dir name in qai_hub_models +id: mediapipe_face_quantized +status: public +headline: Detect faces and locate facial features in real-time video and image streams. +domain: Computer Vision +description: Designed for sub-millisecond processing, this model predicts bounding + boxes and pose skeletons (left eye, right eye, nose tip, mouth, left eye tragion, + and right eye tragion) of faces in an image. +use_case: Object Detection +tags: + - real-time + - quantized +research_paper: https://arxiv.org/abs/1907.05047 +research_paper_title: 'BlazeFace: Sub-millisecond Neural Face Detection on Mobile + GPUs' +license: https://github.com/zmurez/MediaPipePyTorch/blob/master/LICENSE +deploy_license: + https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf +source_repo: https://github.com/zmurez/MediaPipePyTorch/ +technical_details: + Input resolution: 256x256 + Number of output classes: 6 + Number of parameters (MediaPipeFaceDetector): 135K + Model size (MediaPipeFaceDetector): 255 KB + Number of parameters (MediaPipeFaceLandmarkDetector): 603K + Model size (MediaPipeFaceLandmarkDetector): 746 KB +applicable_scenarios: + - Accessibility + - Augmented Reality + - Gaming +related_models: + - mediapipe_face + - mediapipe_pose + - mediapipe_selfie +form_factors: + - Phone + - Tablet + - IoT +has_static_banner: true +has_animated_banner: true +license_type: apache-2.0 +deploy_license_type: AI Model Hub License +dataset: [] diff --git a/qai_hub_models/models/mediapipe_face_quantized/model.py b/qai_hub_models/models/mediapipe_face_quantized/model.py new file mode 100644 index 00000000..555d13da --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/model.py @@ -0,0 +1,118 @@ +# --------------------------------------------------------------------- +# Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# --------------------------------------------------------------------- +from __future__ import annotations + +# isort: off +# This verifies aimet is installed, and this must be included first. +from qai_hub_models.utils.quantization_aimet import ( + AIMETQuantizableMixin, + constrain_quantized_inputs_to_image_range, + tie_observers, +) + +# isort: on + +import torch +from aimet_torch.cross_layer_equalization import equalize_model +from aimet_torch.model_preparer import prepare_model +from aimet_torch.quantsim import QuantizationSimModel, load_encodings_to_sim + +from qai_hub_models.models.mediapipe_face.model import ( + FaceDetector, + FaceLandmarkDetector, + MediaPipeFace, +) +from qai_hub_models.utils.aimet.config_loader import get_default_aimet_config +from qai_hub_models.utils.asset_loaders import CachedWebModelAsset + +MODEL_ID = __name__.split(".")[-2] +MODEL_ASSET_VERSION = 2 +DEFAULT_LANDMARK_DETECTOR_ENCODINGS = "landmark_detector_quantized_encodings.json" + +# This encodings file contains manual overrides. +# The final layers output box confidence logits, most of which are extremely negative. +# There are also only a few values that will be positive, which are the ones we care about. +# When calibrating the range, the quantizer clips the range to exclude these positive outliers. +# The range was manually overriden in the encodings file to be [-128, 127] +DEFAULT_FACE_DETECTOR_ENCODINGS = "face_detector_quantized_encodings.json" + + +class MediaPipeFaceQuantizable(MediaPipeFace): + @classmethod + def from_pretrained( + cls, + face_detector_encodings: str | None = "DEFAULT", + landmark_detector_encodings: str | None = "DEFAULT", + ) -> MediaPipeFaceQuantizable: + return cls( + FaceDetectorQuantizable.from_pretrained(face_detector_encodings), + FaceLandmarkDetectorQuantizable.from_pretrained( + landmark_detector_encodings + ), + ) + + +class FaceDetectorQuantizable(AIMETQuantizableMixin, FaceDetector): + def __init__(self, detector: QuantizationSimModel, anchors: torch.Tensor): + FaceDetector.__init__(self, detector.model, anchors) + AIMETQuantizableMixin.__init__(self, detector) + + @classmethod + def from_pretrained(cls, aimet_encodings: str | None = "DEFAULT"): + fp16_model = FaceDetector.from_pretrained() + input_shape = cls.get_input_spec()["image"][0] + model = prepare_model(fp16_model) + equalize_model(model, input_shape) + sim = QuantizationSimModel( + model, + quant_scheme="tf_enhanced", + default_param_bw=8, + default_output_bw=8, + config_file=get_default_aimet_config(), + dummy_input=torch.rand(input_shape), + ) + tie_observers(sim) + constrain_quantized_inputs_to_image_range(sim) + + if aimet_encodings: + if aimet_encodings == "DEFAULT": + aimet_encodings = CachedWebModelAsset.from_asset_store( + MODEL_ID, MODEL_ASSET_VERSION, DEFAULT_FACE_DETECTOR_ENCODINGS + ).fetch() + load_encodings_to_sim(sim, aimet_encodings) + + return cls(sim, fp16_model.anchors) + + +class FaceLandmarkDetectorQuantizable(AIMETQuantizableMixin, FaceLandmarkDetector): + def __init__(self, detector: QuantizationSimModel): + FaceLandmarkDetector.__init__(self, detector.model) + AIMETQuantizableMixin.__init__(self, detector) + + @classmethod + def from_pretrained(cls, aimet_encodings: str | None = "DEFAULT"): + fp16_model = FaceLandmarkDetector.from_pretrained() + input_shape = cls.get_input_spec()["image"][0] + model = prepare_model(fp16_model) + equalize_model(model, input_shape) + sim = QuantizationSimModel( + model, + quant_scheme="tf_enhanced", + default_param_bw=8, + default_output_bw=8, + config_file=get_default_aimet_config(), + dummy_input=torch.rand(input_shape), + ) + tie_observers(sim) + constrain_quantized_inputs_to_image_range(sim) + + if aimet_encodings: + if aimet_encodings == "DEFAULT": + aimet_encodings = CachedWebModelAsset.from_asset_store( + MODEL_ID, MODEL_ASSET_VERSION, DEFAULT_LANDMARK_DETECTOR_ENCODINGS + ).fetch() + load_encodings_to_sim(sim, aimet_encodings) + + return cls(sim) diff --git a/qai_hub_models/models/mediapipe_face_quantized/perf.yaml b/qai_hub_models/models/mediapipe_face_quantized/perf.yaml new file mode 100644 index 00000000..e6fb9725 --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/perf.yaml @@ -0,0 +1,754 @@ +aggregated: + supported_oses: + - Android + supported_devices: + - Samsung Galaxy S24 + - Samsung Galaxy S24 Ultra + - Samsung Galaxy S24+ + - Samsung Galaxy S23 + - Samsung Galaxy S23 Ultra + - Samsung Galaxy S23+ + - Samsung Galaxy S22 5G + - Samsung Galaxy S22 Ultra 5G + - Samsung Galaxy S22+ 5G + - Samsung Galaxy Tab S8 + - Xiaomi 12 + - Xiaomi 12 Pro + - Samsung Galaxy S21 + - Samsung Galaxy S21 Ultra + - Samsung Galaxy S21+ + - Snapdragon X Elite CRD + - QCS6490 (Proxy) + - RB3 Gen 2 (Proxy) + - QCS8250 (Proxy) + - RB5 (Proxy) + - QCS8550 (Proxy) + - SA8775 (Proxy) + - SA8650 (Proxy) + - SA8255 (Proxy) + - QCS8450 (Proxy) + - XR2 Gen 2 (Proxy) + - Google Pixel 5a 5G + - Google Pixel 4 + - Google Pixel 4a + - Google Pixel 3 + - Google Pixel 3a + - Google Pixel 3a XL + supported_chipsets: + - Snapdragon® 8 Gen 3 + - Snapdragon® 8 Gen 2 + - Snapdragon® 8 Gen 1 + - Snapdragon® 888 + - Snapdragon® X Elite + - Qcs6490 + - Qcs8250 + - Qcs8550 + - Sa8775p + - Sa8650p + - Sa8255p + - Qcs8450 +models: +- name: MediaPipeFaceDetector + performance_metrics: + - torchscript_onnx_tflite: + inference_time: 250.0 + throughput: 4000.0 + estimated_peak_memory_range: + min: 12288 + max: 1268272 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: jo5mle9dg + job_status: Passed + torchscript_onnx_qnn: + inference_time: 295.0 + throughput: 3389.830508474576 + estimated_peak_memory_range: + min: 16384 + max: 47234360 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: jmg9q02wp + job_status: Passed + reference_device_info: + name: Samsung Galaxy S23 + os: '13' + form_factor: Phone + os_name: Android + manufacturer: Samsung + chipset: Snapdragon® 8 Gen 2 + timestamp: '2024-08-26T02:02:40Z' + - torchscript_onnx_tflite: + inference_time: 162.0 + throughput: 6172.83950617284 + estimated_peak_memory_range: + min: 12288 + max: 29771232 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: jopr76x0g + job_status: Passed + torchscript_onnx_qnn: + inference_time: 205.0 + throughput: 4878.048780487805 + estimated_peak_memory_range: + min: 0 + max: 16595488 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: jvgdmn4rg + job_status: Passed + reference_device_info: + name: Samsung Galaxy S24 + os: '14' + form_factor: Phone + os_name: Android + manufacturer: Samsung + chipset: Snapdragon® 8 Gen 3 + timestamp: '2024-08-26T02:02:42Z' + - torchscript_onnx_tflite: + inference_time: 258.0 + throughput: 3875.968992248062 + estimated_peak_memory_range: + min: 12288 + max: 16329928 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: jqpyyz88p + job_status: Passed + torchscript_onnx_qnn: + inference_time: 304.0 + throughput: 3289.4736842105262 + estimated_peak_memory_range: + min: 258048 + max: 1692016 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: jo5mlekdg + job_status: Passed + reference_device_info: + name: QCS8550 (Proxy) + os: '12' + form_factor: Iot + os_name: Android + manufacturer: Qualcomm + chipset: Qcs8550 + timestamp: '2024-08-26T02:02:45Z' + - torchscript_onnx_tflite: + inference_time: 299.0 + throughput: 3344.4816053511704 + estimated_peak_memory_range: + min: 0 + max: 30206272 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: j1p8k2jkp + job_status: Passed + torchscript_onnx_qnn: + inference_time: 358.0 + throughput: 2793.2960893854747 + estimated_peak_memory_range: + min: 208896 + max: 18233840 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: jn5qd01ng + job_status: Passed + reference_device_info: + name: QCS8450 (Proxy) + os: '13' + form_factor: Xr + os_name: Android + manufacturer: Qualcomm + chipset: Qcs8450 + timestamp: '2024-08-26T02:02:52Z' + - torchscript_onnx_tflite: + inference_time: 243.0 + throughput: 4115.22633744856 + estimated_peak_memory_range: + min: 16384 + max: 1405296 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: jn5qd04ng + job_status: Passed + torchscript_onnx_qnn: + inference_time: 301.0 + throughput: 3322.2591362126245 + estimated_peak_memory_range: + min: 225280 + max: 1476512 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: jopr76d0g + job_status: Passed + reference_device_info: + name: SA8650 (Proxy) + os: '13' + form_factor: Auto + os_name: Android + manufacturer: Qualcomm + chipset: Sa8650p + timestamp: '2024-08-26T02:02:47Z' + - torchscript_onnx_tflite: + inference_time: 260.0 + throughput: 3846.153846153846 + estimated_peak_memory_range: + min: 40960 + max: 1339968 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: jw5602o65 + job_status: Passed + torchscript_onnx_qnn: + inference_time: 302.0 + throughput: 3311.2582781456954 + estimated_peak_memory_range: + min: 229376 + max: 1507472 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: jqpyyz28p + job_status: Passed + reference_device_info: + name: SA8775 (Proxy) + os: '13' + form_factor: Auto + os_name: Android + manufacturer: Qualcomm + chipset: Sa8775p + timestamp: '2024-08-26T02:02:49Z' + - torchscript_onnx_tflite: + inference_time: 261.0 + throughput: 3831.417624521073 + estimated_peak_memory_range: + min: 40960 + max: 73004728 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: jwgo9zdqg + job_status: Passed + torchscript_onnx_qnn: + inference_time: 297.0 + throughput: 3367.003367003367 + estimated_peak_memory_range: + min: 249856 + max: 1914304 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: j1p8k2rkp + job_status: Passed + reference_device_info: + name: SA8255 (Proxy) + os: '13' + form_factor: Auto + os_name: Android + manufacturer: Qualcomm + chipset: Sa8255p + timestamp: '2024-08-26T02:02:51Z' + - torchscript_onnx_tflite: + inference_time: 837.0 + throughput: 1194.7431302270013 + estimated_peak_memory_range: + min: 12288 + max: 22755824 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: j7gj8dyv5 + job_status: Passed + torchscript_onnx_qnn: + inference_time: 746.0 + throughput: 1340.4825737265417 + estimated_peak_memory_range: + min: 20480 + max: 8208416 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: jw5602m65 + job_status: Passed + reference_device_info: + name: RB3 Gen 2 (Proxy) + os: '12' + form_factor: Iot + os_name: Android + manufacturer: Qualcomm + chipset: Qcs6490 + timestamp: '2024-08-26T02:02:54Z' + - torchscript_onnx_tflite: + inference_time: 4856.0 + throughput: 205.9308072487644 + estimated_peak_memory_range: + min: 12288 + max: 6768776 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 114 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 114 + job_id: jygz02yo5 + job_status: Passed + reference_device_info: + name: RB5 (Proxy) + os: '12' + form_factor: Iot + os_name: Android + manufacturer: Qualcomm + chipset: Qcs8250 + timestamp: '2024-08-26T02:02:38Z' + - torchscript_onnx_qnn: + inference_time: 415.0 + throughput: 2409.6385542168673 + estimated_peak_memory_range: + min: 552960 + max: 552960 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 118 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 118 + job_id: jqp42n48g + job_status: Passed + reference_device_info: + name: Snapdragon X Elite CRD + os: '11' + form_factor: Compute + os_name: Windows + manufacturer: Qualcomm + chipset: Snapdragon® X Elite + timestamp: '2024-08-26T02:02:43Z' +- name: MediaPipeFaceLandmarkDetector + performance_metrics: + - torchscript_onnx_tflite: + inference_time: 153.0 + throughput: 6535.9477124183 + estimated_peak_memory_range: + min: 16384 + max: 37964312 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: jegnw01kg + job_status: Passed + torchscript_onnx_qnn: + inference_time: 208.0 + throughput: 4807.692307692308 + estimated_peak_memory_range: + min: 24576 + max: 10206680 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: jnp1m218p + job_status: Passed + reference_device_info: + name: Samsung Galaxy S23 + os: '13' + form_factor: Phone + os_name: Android + manufacturer: Samsung + chipset: Snapdragon® 8 Gen 2 + timestamp: '2024-08-26T02:02:41Z' + - torchscript_onnx_tflite: + inference_time: 119.0 + throughput: 8403.361344537816 + estimated_peak_memory_range: + min: 12288 + max: 25933216 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: jep2zxorp + job_status: Passed + torchscript_onnx_qnn: + inference_time: 151.0 + throughput: 6622.516556291391 + estimated_peak_memory_range: + min: 126976 + max: 13670128 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: jz5782nvp + job_status: Passed + reference_device_info: + name: Samsung Galaxy S24 + os: '14' + form_factor: Phone + os_name: Android + manufacturer: Samsung + chipset: Snapdragon® 8 Gen 3 + timestamp: '2024-08-26T02:02:42Z' + - torchscript_onnx_tflite: + inference_time: 164.0 + throughput: 6097.5609756097565 + estimated_peak_memory_range: + min: 12288 + max: 18620584 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: j2p0x4o9p + job_status: Passed + torchscript_onnx_qnn: + inference_time: 204.0 + throughput: 4901.9607843137255 + estimated_peak_memory_range: + min: 135168 + max: 1755624 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: jegnw0qkg + job_status: Passed + reference_device_info: + name: QCS8550 (Proxy) + os: '12' + form_factor: Iot + os_name: Android + manufacturer: Qualcomm + chipset: Qcs8550 + timestamp: '2024-08-26T02:02:46Z' + - torchscript_onnx_tflite: + inference_time: 200.0 + throughput: 5000.0 + estimated_peak_memory_range: + min: 12288 + max: 27271328 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: jogkkv6wg + job_status: Passed + torchscript_onnx_qnn: + inference_time: 249.0 + throughput: 4016.0642570281125 + estimated_peak_memory_range: + min: 126976 + max: 14725648 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: j1glq48jp + job_status: Passed + reference_device_info: + name: QCS8450 (Proxy) + os: '13' + form_factor: Xr + os_name: Android + manufacturer: Qualcomm + chipset: Qcs8450 + timestamp: '2024-08-26T02:02:53Z' + - torchscript_onnx_tflite: + inference_time: 158.0 + throughput: 6329.113924050633 + estimated_peak_memory_range: + min: 12288 + max: 1395968 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: j1glq4wjp + job_status: Passed + torchscript_onnx_qnn: + inference_time: 214.0 + throughput: 4672.897196261682 + estimated_peak_memory_range: + min: 147456 + max: 1450120 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: jep2zxdrp + job_status: Passed + reference_device_info: + name: SA8650 (Proxy) + os: '13' + form_factor: Auto + os_name: Android + manufacturer: Qualcomm + chipset: Sa8650p + timestamp: '2024-08-26T02:02:47Z' + - torchscript_onnx_tflite: + inference_time: 170.0 + throughput: 5882.35294117647 + estimated_peak_memory_range: + min: 16384 + max: 2592072 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: j1p3rno3p + job_status: Passed + torchscript_onnx_qnn: + inference_time: 204.0 + throughput: 4901.9607843137255 + estimated_peak_memory_range: + min: 135168 + max: 1473944 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: j2p0x499p + job_status: Passed + reference_device_info: + name: SA8775 (Proxy) + os: '13' + form_factor: Auto + os_name: Android + manufacturer: Qualcomm + chipset: Sa8775p + timestamp: '2024-08-26T02:02:49Z' + - torchscript_onnx_tflite: + inference_time: 159.0 + throughput: 6289.308176100629 + estimated_peak_memory_range: + min: 12288 + max: 1626504 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: j1pvnqmkg + job_status: Passed + torchscript_onnx_qnn: + inference_time: 207.0 + throughput: 4830.917874396136 + estimated_peak_memory_range: + min: 139264 + max: 1578384 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: jogkkv0wg + job_status: Passed + reference_device_info: + name: SA8255 (Proxy) + os: '13' + form_factor: Auto + os_name: Android + manufacturer: Qualcomm + chipset: Sa8255p + timestamp: '2024-08-26T02:02:51Z' + - torchscript_onnx_tflite: + inference_time: 398.0 + throughput: 2512.5628140703516 + estimated_peak_memory_range: + min: 16384 + max: 19121984 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: jlpenoxop + job_status: Passed + torchscript_onnx_qnn: + inference_time: 479.0 + throughput: 2087.6826722338205 + estimated_peak_memory_range: + min: 12288 + max: 8710224 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: j1p3rn73p + job_status: Passed + reference_device_info: + name: RB3 Gen 2 (Proxy) + os: '12' + form_factor: Iot + os_name: Android + manufacturer: Qualcomm + chipset: Qcs6490 + timestamp: '2024-08-26T02:02:54Z' + - torchscript_onnx_tflite: + inference_time: 2580.0 + throughput: 387.5968992248062 + estimated_peak_memory_range: + min: 16384 + max: 9781408 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 102 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 102 + job_id: jz5wrwz3p + job_status: Passed + reference_device_info: + name: RB5 (Proxy) + os: '12' + form_factor: Iot + os_name: Android + manufacturer: Qualcomm + chipset: Qcs8250 + timestamp: '2024-08-26T02:02:39Z' + - torchscript_onnx_qnn: + inference_time: 376.0 + throughput: 2659.574468085106 + estimated_peak_memory_range: + min: 630784 + max: 630784 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 112 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 112 + job_id: j0pxz9r35 + job_status: Passed + reference_device_info: + name: Snapdragon X Elite CRD + os: '11' + form_factor: Compute + os_name: Windows + manufacturer: Qualcomm + chipset: Snapdragon® X Elite + timestamp: '2024-08-26T02:02:44Z' diff --git a/qai_hub_models/models/mediapipe_face_quantized/test.py b/qai_hub_models/models/mediapipe_face_quantized/test.py new file mode 100644 index 00000000..7df75665 --- /dev/null +++ b/qai_hub_models/models/mediapipe_face_quantized/test.py @@ -0,0 +1,42 @@ +# --------------------------------------------------------------------- +# Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# --------------------------------------------------------------------- +import numpy as np + +from qai_hub_models.models.mediapipe_face.app import MediaPipeFaceApp +from qai_hub_models.models.mediapipe_face.demo import INPUT_IMAGE_ADDRESS +from qai_hub_models.models.mediapipe_face_quantized.demo import main as demo_main +from qai_hub_models.models.mediapipe_face_quantized.model import ( + MODEL_ASSET_VERSION, + MODEL_ID, + MediaPipeFaceQuantizable, +) +from qai_hub_models.utils.asset_loaders import ( + CachedWebModelAsset, + load_image, + load_numpy, +) +from qai_hub_models.utils.testing import skip_clone_repo_check + +GT_LANDMARKS = CachedWebModelAsset.from_asset_store( + MODEL_ID, MODEL_ASSET_VERSION, "expected_landmarks.npy" +) + + +@skip_clone_repo_check +def test_face_app(): + input = load_image(INPUT_IMAGE_ADDRESS) + expected_output = load_numpy(GT_LANDMARKS) + expected_coords, expected_conf = np.split(expected_output, [2], axis=2) + app = MediaPipeFaceApp(MediaPipeFaceQuantizable.from_pretrained()) + result = app.predict_landmarks_from_image(input, raw_output=True) + landmarks = result[3][0] + coords, conf = np.split(landmarks, [2], axis=2) + np.testing.assert_allclose(coords, expected_coords, atol=15) + np.testing.assert_allclose(conf, expected_conf, atol=0.05) + + +@skip_clone_repo_check +def test_demo(): + demo_main(is_test=True) diff --git a/qai_hub_models/models/mediapipe_hand/export.py b/qai_hub_models/models/mediapipe_hand/export.py index 9a63ac27..316e30b7 100644 --- a/qai_hub_models/models/mediapipe_hand/export.py +++ b/qai_hub_models/models/mediapipe_hand/export.py @@ -128,8 +128,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/mediapipe_hand/perf.yaml b/qai_hub_models/models/mediapipe_hand/perf.yaml index e51ddaf4..1d88c95f 100644 --- a/qai_hub_models/models/mediapipe_hand/perf.yaml +++ b/qai_hub_models/models/mediapipe_hand/perf.yaml @@ -44,27 +44,12 @@ aggregated: models: - name: MediaPipeHandDetector performance_metrics: - - torchscript_onnx_tflite: - inference_time: 753.0 - throughput: 1328.0212483399735 - estimated_peak_memory_range: - min: 12288 - max: 123648328 - primary_compute_unit: NPU - precision: fp16 - layer_info: - layers_on_npu: 149 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 149 - job_id: jz57ondvg - job_status: Passed - torchscript_onnx_qnn: + - torchscript_onnx_qnn: inference_time: 791.0 throughput: 1264.2225031605562 estimated_peak_memory_range: - min: 16384 - max: 20928128 + min: 77824 + max: 21048784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,22 +57,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 195 - job_id: j1p3o7m3p - job_status: Passed - torchscript_onnx: - inference_time: 1191.0 - throughput: 839.6305625524769 - estimated_peak_memory_range: - min: 24576 - max: 6196808 - primary_compute_unit: NPU - precision: fp16 - layer_info: - layers_on_npu: 196 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 196 - job_id: jep2od7rg + job_id: jlpen4q1p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +66,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:22:23Z' + timestamp: '2024-08-27T00:36:15Z' - torchscript_onnx_tflite: - inference_time: 534.0 - throughput: 1872.6591760299625 + inference_time: 516.0 + throughput: 1937.984496124031 estimated_peak_memory_range: - min: 12288 - max: 58504304 + min: 16384 + max: 58999792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +80,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: j0px0r13p + job_id: jep2z314p job_status: Passed torchscript_onnx_qnn: - inference_time: 581.0 - throughput: 1721.170395869191 + inference_time: 580.0 + throughput: 1724.1379310344828 estimated_peak_memory_range: min: 0 - max: 19703040 + max: 20348384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +95,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 195 - job_id: j1pv2m1kg + job_id: jz5wrmj6p job_status: Passed torchscript_onnx: - inference_time: 867.0 - throughput: 1153.4025374855826 + inference_time: 865.0 + throughput: 1156.0693641618498 estimated_peak_memory_range: - min: 0 - max: 66334208 + min: 319488 + max: 67423152 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 196 - job_id: j2p0o919p + job_id: j2p0xen9p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +119,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:22:25Z' + timestamp: '2024-08-27T00:36:31Z' - torchscript_onnx_tflite: - inference_time: 739.0 - throughput: 1353.1799729364006 + inference_time: 703.0 + throughput: 1422.475106685633 estimated_peak_memory_range: - min: 12288 - max: 5702240 + min: 24576 + max: 22975080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +133,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jegn1qekp + job_id: j2p0xew6p job_status: Passed torchscript_onnx_qnn: - inference_time: 788.0 - throughput: 1269.0355329949239 + inference_time: 775.0 + throughput: 1290.3225806451612 estimated_peak_memory_range: - min: 815104 - max: 2585896 + min: 819200 + max: 2661024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +148,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 195 - job_id: jz5wyzd3g + job_id: jz5wrmj3p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +157,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:22:13Z' + timestamp: '2024-08-27T00:36:20Z' - torchscript_onnx_tflite: - inference_time: 1331.0 - throughput: 751.3148009015778 + inference_time: 1277.0 + throughput: 783.0853563038371 estimated_peak_memory_range: - min: 12288 - max: 53600944 + min: 20480 + max: 53093520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +171,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jep2odmrg + job_id: jogkkr12g job_status: Passed torchscript_onnx_qnn: - inference_time: 1395.0 - throughput: 716.8458781362007 + inference_time: 1418.0 + throughput: 705.2186177715091 estimated_peak_memory_range: - min: 823296 - max: 18846288 + min: 802816 + max: 16891072 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +186,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 195 - job_id: jegn1q9kp + job_id: jegnwrjkg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +195,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:22:21Z' + timestamp: '2024-08-27T00:36:28Z' - torchscript_onnx_tflite: - inference_time: 748.0 - throughput: 1336.8983957219252 + inference_time: 708.0 + throughput: 1412.4293785310736 estimated_peak_memory_range: - min: 24576 - max: 2834864 + min: 12288 + max: 4252944 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +209,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: j2p0o9r9p + job_id: j1glqed8p job_status: Passed torchscript_onnx_qnn: - inference_time: 802.0 - throughput: 1246.8827930174564 + inference_time: 790.0 + throughput: 1265.8227848101267 estimated_peak_memory_range: - min: 819200 - max: 2115544 + min: 831488 + max: 2413488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +224,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 195 - job_id: jnp1o1d85 + job_id: jnp1mqr8p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +233,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:22:15Z' + timestamp: '2024-08-27T00:36:22Z' - torchscript_onnx_tflite: - inference_time: 755.0 - throughput: 1324.5033112582782 + inference_time: 707.0 + throughput: 1414.4271570014143 estimated_peak_memory_range: min: 24576 - max: 4328176 + max: 4319408 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +247,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: jogk60yw5 + job_id: j1p3rqdlp job_status: Passed torchscript_onnx_qnn: - inference_time: 794.0 - throughput: 1259.4458438287154 + inference_time: 785.0 + throughput: 1273.8853503184714 estimated_peak_memory_range: - min: 815104 - max: 2594968 + min: 831488 + max: 2183160 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +262,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 195 - job_id: jz57onjvg + job_id: jz578vqvp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +271,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:22:17Z' + timestamp: '2024-08-27T00:36:24Z' - torchscript_onnx_tflite: - inference_time: 757.0 - throughput: 1321.003963011889 + inference_time: 706.0 + throughput: 1416.4305949008499 estimated_peak_memory_range: - min: 45056 - max: 79953912 + min: 20480 + max: 4923648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +285,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 149 - job_id: j1glw8kjp + job_id: j1pvnz8jg job_status: Passed torchscript_onnx_qnn: - inference_time: 807.0 - throughput: 1239.1573729863692 + inference_time: 792.0 + throughput: 1262.6262626262626 estimated_peak_memory_range: min: 823296 - max: 2709600 + max: 2468240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +300,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 195 - job_id: j0px0r73p + job_id: j0pxzew35 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +309,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:22:19Z' + timestamp: '2024-08-27T00:36:26Z' - torchscript_onnx_qnn: - inference_time: 957.0 - throughput: 1044.932079414838 + inference_time: 1097.0 + throughput: 911.5770282588878 estimated_peak_memory_range: - min: 786432 - max: 786432 + min: 888832 + max: 888832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +323,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 195 - job_id: jlpe6xrog + job_id: jnp1mqr2p job_status: Passed torchscript_onnx: - inference_time: 1246.0 - throughput: 802.5682182985554 + inference_time: 1213.0 + throughput: 824.4023083264633 estimated_peak_memory_range: - min: 4968448 - max: 4968448 + min: 5021696 + max: 5021696 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +338,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 196 - job_id: jogk60lw5 + job_id: jogkkrjwg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,15 +347,15 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:22:27Z' + timestamp: '2024-08-27T00:36:32Z' - name: MediaPipeHandLandmarkDetector performance_metrics: - torchscript_onnx_tflite: - inference_time: 1010.0 - throughput: 990.0990099009902 + inference_time: 1013.0 + throughput: 987.1668311944719 estimated_peak_memory_range: - min: 32768 - max: 2272904 + min: 16384 + max: 10177144 primary_compute_unit: NPU precision: fp16 layer_info: @@ -393,14 +363,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jqp4e4w8g + job_id: jopr71r9g job_status: Passed torchscript_onnx_qnn: - inference_time: 1099.0 - throughput: 909.9181073703367 + inference_time: 1155.0 + throughput: 865.8008658008658 estimated_peak_memory_range: - min: 708608 - max: 52122656 + min: 782336 + max: 52918848 primary_compute_unit: NPU precision: fp16 layer_info: @@ -408,22 +378,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 208 - job_id: jwgodwvq5 - job_status: Passed - torchscript_onnx: - inference_time: 1548.0 - throughput: 645.9948320413437 - estimated_peak_memory_range: - min: 16384 - max: 8345992 - primary_compute_unit: NPU - precision: fp16 - layer_info: - layers_on_npu: 209 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 209 - job_id: jqpy8248g + job_id: jygz0v6k5 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -432,13 +387,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:22:24Z' + timestamp: '2024-08-27T00:36:15Z' - torchscript_onnx_tflite: - inference_time: 739.0 - throughput: 1353.1799729364006 + inference_time: 730.0 + throughput: 1369.86301369863 estimated_peak_memory_range: - min: 12288 - max: 62396704 + min: 16384 + max: 62685776 primary_compute_unit: NPU precision: fp16 layer_info: @@ -446,14 +401,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jo5m9kzdg + job_id: jqpyyvl7p job_status: Passed torchscript_onnx_qnn: - inference_time: 830.0 - throughput: 1204.8192771084337 + inference_time: 829.0 + throughput: 1206.2726176115802 estimated_peak_memory_range: - min: 0 - max: 20413408 + min: 802816 + max: 18546720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -461,14 +416,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 208 - job_id: j7gj3y0vp + job_id: jmg9q96lp job_status: Passed torchscript_onnx: - inference_time: 1165.0 - throughput: 858.3690987124463 + inference_time: 1132.0 + throughput: 883.3922261484099 estimated_peak_memory_range: min: 0 - max: 63973056 + max: 64182384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -476,7 +431,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 209 - job_id: j1p8jr3k5 + job_id: j1p8kwlkp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -485,13 +440,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:22:26Z' + timestamp: '2024-08-27T00:36:31Z' - torchscript_onnx_tflite: - inference_time: 1004.0 - throughput: 996.01593625498 + inference_time: 991.0 + throughput: 1009.0817356205853 estimated_peak_memory_range: - min: 16384 - max: 1307664 + min: 24576 + max: 188645896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -499,14 +454,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: joprxdy0p + job_id: j1p8kwnxp job_status: Passed torchscript_onnx_qnn: - inference_time: 1100.0 - throughput: 909.0909090909091 + inference_time: 1136.0 + throughput: 880.2816901408451 estimated_peak_memory_range: - min: 835584 - max: 2120200 + min: 819200 + max: 2428456 primary_compute_unit: NPU precision: fp16 layer_info: @@ -514,7 +469,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 208 - job_id: jmg9o23wg + job_id: jmg9q96wp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -523,13 +478,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:22:13Z' + timestamp: '2024-08-27T00:36:21Z' - torchscript_onnx_tflite: - inference_time: 2517.0 - throughput: 397.29837107667856 + inference_time: 2576.0 + throughput: 388.19875776397515 estimated_peak_memory_range: min: 12288 - max: 56324272 + max: 57351424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -537,7 +492,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jqpy82d8g + job_id: jn5qd9n4g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -546,13 +501,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:21:59Z' + timestamp: '2024-08-27T00:36:08Z' - torchscript_onnx_tflite: - inference_time: 1009.0 - throughput: 991.0802775024777 + inference_time: 1023.0 + throughput: 977.5171065493646 estimated_peak_memory_range: - min: 28672 - max: 178245288 + min: 20480 + max: 1597504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -560,14 +515,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: j1p8jr7k5 + job_id: jw560qx05 job_status: Passed torchscript_onnx_qnn: - inference_time: 1079.0 - throughput: 926.7840593141798 + inference_time: 1190.0 + throughput: 840.3361344537815 estimated_peak_memory_range: - min: 827392 - max: 2442304 + min: 831488 + max: 2112224 primary_compute_unit: NPU precision: fp16 layer_info: @@ -575,7 +530,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 208 - job_id: jvgd64rrp + job_id: jvgdm7jrg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -584,13 +539,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:22:15Z' + timestamp: '2024-08-27T00:36:22Z' - torchscript_onnx_tflite: - inference_time: 1012.0 - throughput: 988.1422924901186 + inference_time: 1009.0 + throughput: 991.0802775024777 estimated_peak_memory_range: min: 32768 - max: 188060224 + max: 179089600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -598,14 +553,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jn5q412ng + job_id: jwgo9exxg job_status: Passed torchscript_onnx_qnn: - inference_time: 1085.0 - throughput: 921.6589861751152 + inference_time: 1081.0 + throughput: 925.0693802035153 estimated_peak_memory_range: - min: 819200 - max: 1975640 + min: 811008 + max: 1981104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -613,7 +568,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 208 - job_id: jqp4e4x8g + job_id: jqp42jz8g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -622,13 +577,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:22:17Z' + timestamp: '2024-08-27T00:36:24Z' - torchscript_onnx_tflite: - inference_time: 1005.0 - throughput: 995.0248756218906 + inference_time: 1023.0 + throughput: 977.5171065493646 estimated_peak_memory_range: - min: 12288 - max: 1577256 + min: 24576 + max: 1389776 primary_compute_unit: NPU precision: fp16 layer_info: @@ -636,14 +591,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jw56om165 + job_id: j7gj8k9x5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1107.0 - throughput: 903.342366757001 + inference_time: 1104.0 + throughput: 905.7971014492754 estimated_peak_memory_range: - min: 847872 - max: 2129096 + min: 831488 + max: 2014536 primary_compute_unit: NPU precision: fp16 layer_info: @@ -651,7 +606,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 208 - job_id: jo5m9kwdg + job_id: jo5mlvjdg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -660,10 +615,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:22:20Z' + timestamp: '2024-08-27T00:36:26Z' - torchscript_onnx_qnn: - inference_time: 1302.0 - throughput: 768.0491551459294 + inference_time: 1354.0 + throughput: 738.5524372230428 estimated_peak_memory_range: min: 786432 max: 786432 @@ -674,14 +629,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 208 - job_id: jygzzyxog + job_id: jvgdm7jeg job_status: Passed torchscript_onnx: - inference_time: 1636.0 - throughput: 611.2469437652812 + inference_time: 1591.0 + throughput: 628.5355122564425 estimated_peak_memory_range: - min: 9064448 - max: 9064448 + min: 7467008 + max: 7467008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -689,7 +644,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 209 - job_id: jn5q417ng + job_id: jn5qd9jng job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -698,4 +653,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:22:28Z' + timestamp: '2024-08-27T00:36:33Z' diff --git a/qai_hub_models/models/mediapipe_pose/export.py b/qai_hub_models/models/mediapipe_pose/export.py index 256ec258..8e5d3ca8 100644 --- a/qai_hub_models/models/mediapipe_pose/export.py +++ b/qai_hub_models/models/mediapipe_pose/export.py @@ -128,8 +128,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/mediapipe_pose/perf.yaml b/qai_hub_models/models/mediapipe_pose/perf.yaml index 7ec88ba1..f953077f 100644 --- a/qai_hub_models/models/mediapipe_pose/perf.yaml +++ b/qai_hub_models/models/mediapipe_pose/perf.yaml @@ -45,11 +45,11 @@ models: - name: MediaPipePoseDetector performance_metrics: - torchscript_onnx_tflite: - inference_time: 793.0 - throughput: 1261.034047919294 + inference_time: 771.0 + throughput: 1297.0168612191958 estimated_peak_memory_range: - min: 36864 - max: 14530304 + min: 28672 + max: 114921912 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: j2p0o926p + job_id: jqpyyvy7p job_status: Passed torchscript_onnx_qnn: - inference_time: 851.0 - throughput: 1175.0881316098707 + inference_time: 836.0 + throughput: 1196.1722488038276 estimated_peak_memory_range: min: 12288 - max: 106965760 + max: 108545720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jnp1o1e25 + job_id: jnp1mqk2p job_status: Passed torchscript_onnx: - inference_time: 993.0 - throughput: 1007.0493454179255 + inference_time: 1003.0 + throughput: 997.0089730807578 estimated_peak_memory_range: - min: 16384 - max: 4035368 + min: 12288 + max: 4120496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 139 - job_id: jogk603w5 + job_id: j1p3rqllp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:21:19Z' + timestamp: '2024-08-27T00:35:32Z' - torchscript_onnx_tflite: - inference_time: 568.0 - throughput: 1760.5633802816901 + inference_time: 563.0 + throughput: 1776.1989342806394 estimated_peak_memory_range: - min: 16384 - max: 45314560 + min: 61440 + max: 45977264 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: jogk60q25 + job_id: j1p8kwkxp job_status: Passed torchscript_onnx_qnn: - inference_time: 609.0 - throughput: 1642.0361247947455 + inference_time: 605.0 + throughput: 1652.892561983471 estimated_peak_memory_range: - min: 208896 - max: 15006272 + min: 212992 + max: 17387776 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jz5wyzq3g + job_id: jz578v1lp job_status: Passed torchscript_onnx: - inference_time: 761.0 - throughput: 1314.060446780552 + inference_time: 744.0 + throughput: 1344.0860215053763 estimated_peak_memory_range: min: 0 - max: 48946576 + max: 49621872 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 139 - job_id: j1glw83jp + job_id: j1pvnzyjg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:21:21Z' + timestamp: '2024-08-27T00:35:34Z' - torchscript_onnx_tflite: - inference_time: 810.0 - throughput: 1234.567901234568 + inference_time: 769.0 + throughput: 1300.3901170351105 estimated_peak_memory_range: min: 12288 - max: 1312984 + max: 1369736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: j1glw828p + job_id: jn5qd9d4g job_status: Passed torchscript_onnx_qnn: - inference_time: 828.0 - throughput: 1207.729468599034 + inference_time: 819.0 + throughput: 1221.001221001221 estimated_peak_memory_range: - min: 221184 - max: 1641336 + min: 241664 + max: 1416136 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jz57onxvg + job_id: jegnwrdrg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:21:08Z' + timestamp: '2024-08-27T00:35:23Z' - torchscript_onnx_tflite: - inference_time: 1926.0 - throughput: 519.2107995846313 + inference_time: 1897.0 + throughput: 527.1481286241434 estimated_peak_memory_range: - min: 61440 - max: 42246432 + min: 65536 + max: 42310208 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: j1p3o71lp + job_id: jw560q005 job_status: Passed torchscript_onnx_qnn: - inference_time: 1986.0 - throughput: 503.5246727089627 + inference_time: 1988.0 + throughput: 503.01810865191146 estimated_peak_memory_range: min: 208896 - max: 13669984 + max: 15557008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: j2p0o9l9p + job_id: j1glqe98p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:21:17Z' + timestamp: '2024-08-27T00:35:31Z' - torchscript_onnx_tflite: - inference_time: 812.0 - throughput: 1231.527093596059 + inference_time: 770.0 + throughput: 1298.7012987012988 estimated_peak_memory_range: min: 24576 - max: 5916760 + max: 3952424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: j1pv2mrjg + job_id: jwgo9e9xg job_status: Passed torchscript_onnx_qnn: - inference_time: 829.0 - throughput: 1206.2726176115802 + inference_time: 818.0 + throughput: 1222.4938875305625 estimated_peak_memory_range: - min: 229376 - max: 1472896 + min: 221184 + max: 1911720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: j0px0ry3p + job_id: jep2z3q4p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:21:10Z' + timestamp: '2024-08-27T00:35:25Z' - torchscript_onnx_tflite: - inference_time: 796.0 - throughput: 1256.2814070351758 + inference_time: 774.0 + throughput: 1291.9896640826873 estimated_peak_memory_range: - min: 28672 - max: 1438000 + min: 32768 + max: 9181920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: jlpe6xw1g + job_id: j7gj8k8x5 job_status: Passed torchscript_onnx_qnn: - inference_time: 830.0 - throughput: 1204.8192771084337 + inference_time: 820.0 + throughput: 1219.5121951219512 estimated_peak_memory_range: min: 229376 - max: 1503440 + max: 1960512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jegn1q3kp + job_id: j2p0xe86p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:21:13Z' + timestamp: '2024-08-27T00:35:27Z' - torchscript_onnx_tflite: - inference_time: 793.0 - throughput: 1261.034047919294 + inference_time: 769.0 + throughput: 1300.3901170351105 estimated_peak_memory_range: - min: 12288 - max: 14407224 + min: 24576 + max: 1602744 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 106 - job_id: jz5wyzq6g + job_id: jz5wrm06p job_status: Passed torchscript_onnx_qnn: - inference_time: 829.0 - throughput: 1206.2726176115802 + inference_time: 827.0 + throughput: 1209.1898428053205 estimated_peak_memory_range: - min: 221184 - max: 1458560 + min: 253952 + max: 1779952 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jep2odlrg + job_id: jogkkrw2g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:21:15Z' + timestamp: '2024-08-27T00:35:29Z' - torchscript_onnx_qnn: - inference_time: 989.0 - throughput: 1011.1223458038422 + inference_time: 955.0 + throughput: 1047.1204188481674 estimated_peak_memory_range: - min: 1687552 - max: 1687552 + min: 438272 + max: 438272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jnp1o1e85 + job_id: j0pxze815 job_status: Passed torchscript_onnx: - inference_time: 1064.0 - throughput: 939.8496240601504 + inference_time: 1059.0 + throughput: 944.2870632672333 estimated_peak_memory_range: - min: 3981312 - max: 3981312 + min: 4526080 + max: 4526080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 139 - job_id: j1p3o7e3p + job_id: jlpen401p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,15 +377,15 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:21:23Z' + timestamp: '2024-08-27T00:35:35Z' - name: MediaPipePoseLandmarkDetector performance_metrics: - torchscript_onnx_tflite: - inference_time: 839.0 - throughput: 1191.8951132300358 + inference_time: 811.0 + throughput: 1233.0456226880394 estimated_peak_memory_range: - min: 57344 - max: 182268800 + min: 12288 + max: 184392864 primary_compute_unit: NPU precision: fp16 layer_info: @@ -393,14 +393,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: j1p8jrmx5 + job_id: j2p0xex6p job_status: Passed torchscript_onnx_qnn: - inference_time: 906.0 - throughput: 1103.7527593818984 + inference_time: 942.0 + throughput: 1061.5711252653928 estimated_peak_memory_range: - min: 176128 - max: 9922688 + min: 16384 + max: 180606864 primary_compute_unit: NPU precision: fp16 layer_info: @@ -408,14 +408,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jvgd64oep + job_id: jvgdm7yeg job_status: Passed torchscript_onnx: - inference_time: 1343.0 - throughput: 744.6016381236038 + inference_time: 1309.0 + throughput: 763.9419404125287 estimated_peak_memory_range: min: 12288 - max: 9589472 + max: 9303312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -423,7 +423,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 291 - job_id: jn5q413ng + job_id: jwgo9e7xg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -432,13 +432,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:21:20Z' + timestamp: '2024-08-27T00:35:32Z' - torchscript_onnx_tflite: - inference_time: 590.0 - throughput: 1694.915254237288 + inference_time: 600.0 + throughput: 1666.6666666666667 estimated_peak_memory_range: min: 12288 - max: 91339936 + max: 91640640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -446,14 +446,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jn5q41r4g + job_id: jogkkrk2g job_status: Passed torchscript_onnx_qnn: - inference_time: 681.0 - throughput: 1468.4287812041116 + inference_time: 655.0 + throughput: 1526.7175572519084 estimated_peak_memory_range: min: 0 - max: 21188240 + max: 23331808 primary_compute_unit: NPU precision: fp16 layer_info: @@ -461,14 +461,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jmg9o2wwg + job_id: jqp42j6vg job_status: Passed torchscript_onnx: - inference_time: 966.0 - throughput: 1035.1966873706003 + inference_time: 958.0 + throughput: 1043.8413361169103 estimated_peak_memory_range: min: 0 - max: 93794144 + max: 95294752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -476,7 +476,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 291 - job_id: jw56omn65 + job_id: j7gj8k6x5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -485,13 +485,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:21:22Z' + timestamp: '2024-08-27T00:35:34Z' - torchscript_onnx_tflite: - inference_time: 815.0 - throughput: 1226.993865030675 + inference_time: 784.0 + throughput: 1275.5102040816328 estimated_peak_memory_range: - min: 24576 - max: 1501920 + min: 12288 + max: 1463504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -499,14 +499,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jw56omz05 + job_id: j1glqeq8p job_status: Passed torchscript_onnx_qnn: - inference_time: 903.0 - throughput: 1107.4197120708748 + inference_time: 881.0 + throughput: 1135.0737797956867 estimated_peak_memory_range: - min: 823296 - max: 2174952 + min: 819200 + max: 2036272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -514,7 +514,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jqp4e4v8g + job_id: jopr71m9g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -523,13 +523,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:21:09Z' + timestamp: '2024-08-27T00:35:23Z' - torchscript_onnx_tflite: - inference_time: 1833.0 - throughput: 545.5537370430987 + inference_time: 1721.0 + throughput: 581.0575246949448 estimated_peak_memory_range: min: 12288 - max: 81884480 + max: 80627376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -537,7 +537,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jwgodwnx5 + job_id: j1p3rqrlp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -546,13 +546,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:20:54Z' + timestamp: '2024-08-27T00:35:12Z' - torchscript_onnx_tflite: - inference_time: 808.0 - throughput: 1237.6237623762377 + inference_time: 812.0 + throughput: 1231.527093596059 estimated_peak_memory_range: min: 12288 - max: 1531808 + max: 1414872 primary_compute_unit: NPU precision: fp16 layer_info: @@ -560,14 +560,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: j7gj3y2xp + job_id: j1pvnznjg job_status: Passed torchscript_onnx_qnn: - inference_time: 908.0 - throughput: 1101.3215859030836 + inference_time: 904.0 + throughput: 1106.1946902654868 estimated_peak_memory_range: - min: 823296 - max: 2085128 + min: 811008 + max: 2020288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -575,7 +575,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jo5m9k3dg + job_id: jqpyyvk7p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -584,13 +584,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:21:11Z' + timestamp: '2024-08-27T00:35:26Z' - torchscript_onnx_tflite: - inference_time: 808.0 - throughput: 1237.6237623762377 + inference_time: 803.0 + throughput: 1245.3300124533 estimated_peak_memory_range: min: 24576 - max: 1567256 + max: 2360992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -598,14 +598,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jygzzyjkg + job_id: jlpen4n1p job_status: Passed torchscript_onnx_qnn: - inference_time: 874.0 - throughput: 1144.1647597254005 + inference_time: 883.0 + throughput: 1132.5028312570782 estimated_peak_memory_range: min: 819200 - max: 2070088 + max: 2093088 primary_compute_unit: NPU precision: fp16 layer_info: @@ -613,7 +613,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: joprxde0p + job_id: j1p8kwdxp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -622,13 +622,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:21:13Z' + timestamp: '2024-08-27T00:35:27Z' - torchscript_onnx_tflite: - inference_time: 837.0 - throughput: 1194.7431302270013 + inference_time: 800.0 + throughput: 1250.0 estimated_peak_memory_range: - min: 20480 - max: 1560088 + min: 40960 + max: 194673080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -636,14 +636,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 219 - job_id: jmg9o2wlg + job_id: jmg9q97lp job_status: Passed torchscript_onnx_qnn: - inference_time: 884.0 - throughput: 1131.2217194570135 + inference_time: 878.0 + throughput: 1138.9521640091116 estimated_peak_memory_range: - min: 819200 - max: 2326552 + min: 827392 + max: 2024512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -651,7 +651,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jqpy8268g + job_id: jn5qd9x4g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -660,10 +660,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:21:16Z' + timestamp: '2024-08-27T00:35:29Z' - torchscript_onnx_qnn: - inference_time: 1108.0 - throughput: 902.5270758122743 + inference_time: 1132.0 + throughput: 883.3922261484099 estimated_peak_memory_range: min: 786432 max: 786432 @@ -674,14 +674,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jvgd64orp + job_id: jo5mlv1wg job_status: Passed torchscript_onnx: - inference_time: 1356.0 - throughput: 737.4631268436578 + inference_time: 1363.0 + throughput: 733.6757153338225 estimated_peak_memory_range: - min: 9056256 - max: 9056256 + min: 9281536 + max: 9281536 primary_compute_unit: NPU precision: fp16 layer_info: @@ -689,7 +689,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 291 - job_id: jwgodw3q5 + job_id: jygz0vqk5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -698,4 +698,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:21:24Z' + timestamp: '2024-08-27T00:35:36Z' diff --git a/qai_hub_models/models/mediapipe_selfie/perf.yaml b/qai_hub_models/models/mediapipe_selfie/perf.yaml index 6a23931a..1b791f8f 100644 --- a/qai_hub_models/models/mediapipe_selfie/perf.yaml +++ b/qai_hub_models/models/mediapipe_selfie/perf.yaml @@ -45,11 +45,11 @@ models: - name: MediaPipe-Selfie-Segmentation performance_metrics: - torchscript_onnx_tflite: - inference_time: 727.0 - throughput: 1375.515818431912 + inference_time: 701.0 + throughput: 1426.5335235378031 estimated_peak_memory_range: - min: 12288 - max: 2419920 + min: 0 + max: 1523760 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: jogk60725 + job_id: jqp429zvg job_status: Passed torchscript_onnx_qnn: - inference_time: 786.0 - throughput: 1272.264631043257 + inference_time: 771.0 + throughput: 1297.0168612191958 estimated_peak_memory_range: min: 16384 - max: 67729784 + max: 3407432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: j7gj3yzxp + job_id: j2p0xvn6p job_status: Passed torchscript_onnx: - inference_time: 1363.0 - throughput: 733.6757153338225 + inference_time: 1323.0 + throughput: 755.8578987150415 estimated_peak_memory_range: - min: 352256 - max: 2792280 + min: 319488 + max: 2895512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: jqp4e4dvg + job_id: j1pvn9jjg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:20:22Z' + timestamp: '2024-08-27T11:55:21Z' - torchscript_onnx_tflite: - inference_time: 488.0 - throughput: 2049.1803278688526 + inference_time: 462.0 + throughput: 2164.5021645021643 estimated_peak_memory_range: min: 12288 - max: 27469248 + max: 27763952 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: jn5q41e4g + job_id: j0pxzdw15 job_status: Passed torchscript_onnx_qnn: - inference_time: 513.0 - throughput: 1949.317738791423 + inference_time: 512.0 + throughput: 1953.125 estimated_peak_memory_range: min: 802816 - max: 15447936 + max: 16318528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jlpe6xe1g + job_id: j1p8k4lxp job_status: Passed torchscript_onnx: - inference_time: 895.0 - throughput: 1117.31843575419 + inference_time: 893.0 + throughput: 1119.8208286674133 estimated_peak_memory_range: min: 0 - max: 31320384 + max: 31547168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: j0px0r61p + job_id: j7gj8wjx5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:20:23Z' + timestamp: '2024-08-27T11:55:22Z' - torchscript_onnx_tflite: - inference_time: 733.0 - throughput: 1364.256480218281 + inference_time: 698.0 + throughput: 1432.6647564469913 estimated_peak_memory_range: - min: 20480 - max: 1468608 + min: 12288 + max: 4393832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: j1glw868p + job_id: jo5mldjwg job_status: Passed torchscript_onnx_qnn: - inference_time: 761.0 - throughput: 1314.060446780552 + inference_time: 760.0 + throughput: 1315.7894736842106 estimated_peak_memory_range: min: 823296 - max: 2150024 + max: 2493920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jz5wyz36g + job_id: jn5qdmj4g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:20:16Z' + timestamp: '2024-08-27T11:55:17Z' - torchscript_onnx_tflite: - inference_time: 963.0 - throughput: 1038.4215991692627 + inference_time: 919.0 + throughput: 1088.139281828074 estimated_peak_memory_range: - min: 12288 - max: 27887248 + min: 16384 + max: 28113328 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: jw56ome05 + job_id: jegnw7jrg job_status: Passed torchscript_onnx_qnn: - inference_time: 989.0 - throughput: 1011.1223458038422 + inference_time: 985.0 + throughput: 1015.2284263959391 estimated_peak_memory_range: min: 802816 - max: 16703600 + max: 15570240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jz57onllg + job_id: jwgo94jxg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:20:21Z' + timestamp: '2024-08-27T11:55:20Z' - torchscript_onnx_tflite: - inference_time: 727.0 - throughput: 1375.515818431912 + inference_time: 701.0 + throughput: 1426.5335235378031 estimated_peak_memory_range: - min: 24576 - max: 1843608 + min: 12288 + max: 1556512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: j1p3o7vlp + job_id: jopr7nz9g job_status: Passed torchscript_onnx_qnn: - inference_time: 761.0 - throughput: 1314.060446780552 + inference_time: 766.0 + throughput: 1305.4830287206266 estimated_peak_memory_range: min: 819200 - max: 1884400 + max: 2090904 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jmg9o2ylg + job_id: j1glq1j8p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:20:17Z' + timestamp: '2024-08-27T11:55:18Z' - torchscript_onnx_tflite: - inference_time: 734.0 - throughput: 1362.3978201634877 + inference_time: 702.0 + throughput: 1424.5014245014245 estimated_peak_memory_range: min: 12288 - max: 1598136 + max: 1624600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: jwgodwkx5 + job_id: jep2zv24p job_status: Passed torchscript_onnx_qnn: - inference_time: 769.0 - throughput: 1300.3901170351105 + inference_time: 766.0 + throughput: 1305.4830287206266 estimated_peak_memory_range: - min: 823296 - max: 2126320 + min: 811008 + max: 2159792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jnp1o1w25 + job_id: jw560dk05 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:20:18Z' + timestamp: '2024-08-27T11:55:18Z' - torchscript_onnx_tflite: - inference_time: 740.0 - throughput: 1351.3513513513512 + inference_time: 705.0 + throughput: 1418.4397163120568 estimated_peak_memory_range: - min: 20480 - max: 20349880 + min: 12288 + max: 62087528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: j1pv2m0jg + job_id: jqpyy797p job_status: Passed torchscript_onnx_qnn: - inference_time: 768.0 - throughput: 1302.0833333333333 + inference_time: 809.0 + throughput: 1236.0939431396787 estimated_peak_memory_range: min: 819200 - max: 2340112 + max: 2167232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jvgd64qep + job_id: j1p3rwylp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:20:19Z' + timestamp: '2024-08-27T11:55:19Z' - torchscript_onnx_qnn: - inference_time: 906.0 - throughput: 1103.7527593818984 + inference_time: 995.0 + throughput: 1005.0251256281407 estimated_peak_memory_range: min: 786432 max: 786432 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 138 - job_id: jygzzyokg + job_id: jogkk9j2g job_status: Passed torchscript_onnx: - inference_time: 1333.0 - throughput: 750.1875468867216 + inference_time: 1582.0 + throughput: 632.1112515802781 estimated_peak_memory_range: - min: 1982464 - max: 1982464 + min: 1994752 + max: 1994752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 140 - job_id: jo5m9k6wg + job_id: jlpenlj1p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:20:24Z' + timestamp: '2024-08-27T11:55:23Z' diff --git a/qai_hub_models/models/midas/model.py b/qai_hub_models/models/midas/model.py index b6034ad1..123ee925 100644 --- a/qai_hub_models/models/midas/model.py +++ b/qai_hub_models/models/midas/model.py @@ -145,3 +145,7 @@ def forward(self, image): @staticmethod def get_channel_last_inputs() -> List[str]: return ["image"] + + @staticmethod + def get_channel_last_outputs() -> List[str]: + return ["depth_estimates"] diff --git a/qai_hub_models/models/midas/perf.yaml b/qai_hub_models/models/midas/perf.yaml index 04b12c1a..eeda1179 100644 --- a/qai_hub_models/models/midas/perf.yaml +++ b/qai_hub_models/models/midas/perf.yaml @@ -45,41 +45,41 @@ models: - name: Midas-V2 performance_metrics: - torchscript_onnx_tflite: - inference_time: 3434.0 - throughput: 291.20559114735005 + inference_time: 3191.0 + throughput: 313.38138514572233 estimated_peak_memory_range: - min: 16384 - max: 2054936 + min: 12288 + max: 2080240 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 139 + layers_on_npu: 138 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 139 - job_id: j1glw848p + total_layers: 138 + job_id: jo5mld7wg job_status: Passed torchscript_onnx_qnn: - inference_time: 3388.0 - throughput: 295.159386068477 + inference_time: 3295.0 + throughput: 303.49013657056145 estimated_peak_memory_range: - min: 806912 - max: 25779368 + min: 299008 + max: 46963936 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 198 + layers_on_npu: 197 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 198 - job_id: jygzzy2kg + total_layers: 197 + job_id: jogkk912g job_status: Passed torchscript_onnx: - inference_time: 3299.0 - throughput: 303.12215822976657 + inference_time: 3347.0 + throughput: 298.7750224081267 estimated_peak_memory_range: min: 12288 - max: 42647592 + max: 42745104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 199 - job_id: jo5m9kqwg + job_id: jlpenlq1p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,43 +96,43 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:19:43Z' + timestamp: '2024-08-27T11:54:48Z' - torchscript_onnx_tflite: - inference_time: 2405.0 - throughput: 415.8004158004158 + inference_time: 2281.0 + throughput: 438.4042086804033 estimated_peak_memory_range: - min: 16384 - max: 88657456 + min: 12288 + max: 88309664 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 139 + layers_on_npu: 138 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 139 - job_id: jw56om205 + total_layers: 138 + job_id: jegnw74rg job_status: Passed torchscript_onnx_qnn: - inference_time: 2397.0 - throughput: 417.18815185648725 + inference_time: 2299.0 + throughput: 434.97172683775557 estimated_peak_memory_range: - min: 802816 - max: 25016736 + min: 0 + max: 26304320 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 198 + layers_on_npu: 197 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 198 - job_id: jz5wyz26g + total_layers: 197 + job_id: jn5qdmn4g job_status: Passed torchscript_onnx: - inference_time: 2392.0 - throughput: 418.0602006688963 + inference_time: 2386.0 + throughput: 419.11148365465215 estimated_peak_memory_range: min: 0 - max: 89017152 + max: 90310592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 199 - job_id: jegn1qlrp + job_id: jygz046k5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,36 +149,36 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:19:44Z' + timestamp: '2024-08-27T11:54:48Z' - torchscript_onnx_tflite: - inference_time: 3426.0 - throughput: 291.8855808523059 + inference_time: 3186.0 + throughput: 313.8731952291274 estimated_peak_memory_range: - min: 20480 - max: 2084552 + min: 45056 + max: 45850072 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 139 + layers_on_npu: 138 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 139 - job_id: j1p3o7nlp + total_layers: 138 + job_id: jopr7nr9g job_status: Passed torchscript_onnx_qnn: - inference_time: 3167.0 - throughput: 315.75623618566465 + inference_time: 3070.0 + throughput: 325.7328990228013 estimated_peak_memory_range: - min: 823296 - max: 2303912 + min: 819200 + max: 2296344 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 198 + layers_on_npu: 197 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 198 - job_id: jnp1o1y25 + total_layers: 197 + job_id: jw560dx05 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,36 +187,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:19:37Z' + timestamp: '2024-08-27T11:54:43Z' - torchscript_onnx_tflite: - inference_time: 4948.0 - throughput: 202.1018593371059 + inference_time: 4746.0 + throughput: 210.70375052675936 estimated_peak_memory_range: - min: 16384 - max: 92198096 + min: 12288 + max: 91025872 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 139 + layers_on_npu: 138 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 139 - job_id: jwgodwzx5 + total_layers: 138 + job_id: jep2zv14p job_status: Passed torchscript_onnx_qnn: - inference_time: 4851.0 - throughput: 206.14306328592042 + inference_time: 4905.0 + throughput: 203.8735983690112 estimated_peak_memory_range: min: 802816 - max: 27232608 + max: 24217232 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 198 + layers_on_npu: 197 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 198 - job_id: j0px0rn1p + total_layers: 197 + job_id: j7gj8w9x5 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,36 +225,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:19:42Z' + timestamp: '2024-08-27T11:54:47Z' - torchscript_onnx_tflite: - inference_time: 3445.0 - throughput: 290.2757619738752 + inference_time: 3191.0 + throughput: 313.38138514572233 estimated_peak_memory_range: - min: 24576 - max: 2055152 + min: 16384 + max: 1952888 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 139 + layers_on_npu: 138 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 139 - job_id: j1pv2mqjg + total_layers: 138 + job_id: jqpyy7l7p job_status: Passed torchscript_onnx_qnn: - inference_time: 3185.0 - throughput: 313.9717425431711 + inference_time: 3083.0 + throughput: 324.3593902043464 estimated_peak_memory_range: - min: 835584 - max: 2370816 + min: 831488 + max: 2034008 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 198 + layers_on_npu: 197 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 198 - job_id: jvgd64eep + total_layers: 197 + job_id: j1p3rwdlp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,36 +263,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:19:38Z' + timestamp: '2024-08-27T11:54:44Z' - torchscript_onnx_tflite: - inference_time: 3429.0 - throughput: 291.6302128900554 + inference_time: 3190.0 + throughput: 313.47962382445144 estimated_peak_memory_range: - min: 233472 - max: 18897528 + min: 16384 + max: 2268016 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 139 + layers_on_npu: 138 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 139 - job_id: j7gj3ydxp + total_layers: 138 + job_id: j2p0xvw6p job_status: Passed torchscript_onnx_qnn: - inference_time: 3174.0 - throughput: 315.059861373661 + inference_time: 3084.0 + throughput: 324.25421530479895 estimated_peak_memory_range: - min: 823296 - max: 2518392 + min: 835584 + max: 2164544 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 198 + layers_on_npu: 197 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 198 - job_id: jz57on0lg + total_layers: 197 + job_id: jwgo94xxg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,36 +301,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:19:39Z' + timestamp: '2024-08-27T11:54:45Z' - torchscript_onnx_tflite: - inference_time: 3437.0 - throughput: 290.9514111143439 + inference_time: 3208.0 + throughput: 311.7206982543641 estimated_peak_memory_range: - min: 28672 - max: 3284648 + min: 20480 + max: 2321128 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 139 + layers_on_npu: 138 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 139 - job_id: jlpe6xo1g + total_layers: 138 + job_id: j1p8k4nxp job_status: Passed torchscript_onnx_qnn: - inference_time: 3194.0 - throughput: 313.08703819661866 + inference_time: 3105.0 + throughput: 322.061191626409 estimated_peak_memory_range: - min: 823296 - max: 2060920 + min: 4329472 + max: 5539808 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 198 + layers_on_npu: 197 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 198 - job_id: jqp4e4kvg + total_layers: 197 + job_id: j1pvn98jg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,28 +339,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:19:40Z' + timestamp: '2024-08-27T11:54:46Z' - torchscript_onnx_qnn: - inference_time: 3410.0 - throughput: 293.2551319648094 + inference_time: 3281.0 + throughput: 304.7851264858275 estimated_peak_memory_range: min: 786432 max: 786432 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 198 + layers_on_npu: 197 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 198 - job_id: jmg9o2jlg + total_layers: 197 + job_id: j1glq1d8p job_status: Passed torchscript_onnx: - inference_time: 3369.0 - throughput: 296.8239833778569 + inference_time: 3530.0 + throughput: 283.28611898017 estimated_peak_memory_range: - min: 38707200 - max: 38707200 + min: 40275968 + max: 40275968 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 199 - job_id: joprxd89p + job_id: jz5wr1j6p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:19:45Z' + timestamp: '2024-08-27T11:54:49Z' diff --git a/qai_hub_models/models/midas_quantized/perf.yaml b/qai_hub_models/models/midas_quantized/perf.yaml index 81fb16a0..5ce24523 100644 --- a/qai_hub_models/models/midas_quantized/perf.yaml +++ b/qai_hub_models/models/midas_quantized/perf.yaml @@ -48,34 +48,34 @@ models: - name: Midas-V2-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 1157.0 - throughput: 864.304235090752 + inference_time: 1074.0 + throughput: 931.0986964618249 estimated_peak_memory_range: - min: 36864 - max: 9391232 + min: 12288 + max: 248528952 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: j1glw8v8p + total_layers: 145 + job_id: j1glqeylp job_status: Passed torchscript_onnx_qnn: - inference_time: 1494.0 - throughput: 669.3440428380187 + inference_time: 1422.0 + throughput: 703.2348804500704 estimated_peak_memory_range: - min: 151552 - max: 66762832 + min: 217088 + max: 7888920 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: jmg9o20lg + total_layers: 146 + job_id: jmg9q9zvp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -84,36 +84,36 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:18:52Z' + timestamp: '2024-08-27T00:33:22Z' - torchscript_onnx_tflite: - inference_time: 814.0 - throughput: 1228.5012285012285 + inference_time: 761.0 + throughput: 1314.060446780552 estimated_peak_memory_range: min: 12288 - max: 88226592 + max: 90070480 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: jw56omy05 + total_layers: 145 + job_id: jw560q875 job_status: Passed torchscript_onnx_qnn: - inference_time: 1069.0 - throughput: 935.4536950420954 + inference_time: 1015.0 + throughput: 985.2216748768473 estimated_peak_memory_range: - min: 0 - max: 24866416 + min: 208896 + max: 21463616 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: jnp1o1225 + total_layers: 146 + job_id: jnp1mqnlp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -122,36 +122,36 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:18:53Z' + timestamp: '2024-08-27T00:33:23Z' - torchscript_onnx_tflite: - inference_time: 1148.0 - throughput: 871.0801393728223 + inference_time: 1072.0 + throughput: 932.8358208955224 estimated_peak_memory_range: - min: 28672 - max: 247077288 + min: 12288 + max: 248447024 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: j1p3o7jlp + total_layers: 145 + job_id: j1p3rqzzp job_status: Passed torchscript_onnx_qnn: - inference_time: 1349.0 - throughput: 741.2898443291327 + inference_time: 1294.0 + throughput: 772.7975270479135 estimated_peak_memory_range: min: 229376 - max: 1516456 + max: 1817920 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: jz57on2lg + total_layers: 146 + job_id: jz578vorp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -160,36 +160,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:18:55Z' + timestamp: '2024-08-27T00:33:25Z' - torchscript_onnx_tflite: - inference_time: 1554.0 - throughput: 643.5006435006435 + inference_time: 1424.0 + throughput: 702.2471910112359 estimated_peak_memory_range: min: 81920 - max: 87210528 + max: 88805392 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: jwgodw2x5 + total_layers: 145 + job_id: jwgo9eldg job_status: Passed torchscript_onnx_qnn: - inference_time: 1844.0 - throughput: 542.2993492407809 + inference_time: 1759.0 + throughput: 568.5048322910744 estimated_peak_memory_range: - min: 212992 - max: 23037232 + min: 208896 + max: 28436256 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: jegn1q0rp + total_layers: 146 + job_id: jegnwr1mg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -198,36 +198,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:19:00Z' + timestamp: '2024-08-27T00:33:29Z' - torchscript_onnx_tflite: - inference_time: 1153.0 - throughput: 867.3026886383348 + inference_time: 1067.0 + throughput: 937.207122774133 estimated_peak_memory_range: min: 12288 - max: 1830544 + max: 1807472 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: j1pv2m6jg + total_layers: 145 + job_id: j1pvnzlmg job_status: Passed torchscript_onnx_qnn: - inference_time: 1350.0 - throughput: 740.7407407407408 + inference_time: 1312.0 + throughput: 762.1951219512196 estimated_peak_memory_range: - min: 233472 - max: 1913664 + min: 229376 + max: 1899224 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: jqp4e4nvg + total_layers: 146 + job_id: jqp42jelg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -236,36 +236,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:18:57Z' + timestamp: '2024-08-27T00:33:26Z' - torchscript_onnx_tflite: - inference_time: 1150.0 - throughput: 869.5652173913044 + inference_time: 1079.0 + throughput: 926.7840593141798 estimated_peak_memory_range: - min: 24576 - max: 256581256 + min: 12288 + max: 16234408 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: j7gj3yvxp + total_layers: 145 + job_id: j7gj8kr85 job_status: Passed torchscript_onnx_qnn: - inference_time: 1346.0 - throughput: 742.9420505200594 + inference_time: 1313.0 + throughput: 761.6146230007616 estimated_peak_memory_range: - min: 225280 - max: 1480672 + min: 221184 + max: 2072528 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: j0px0r91p + total_layers: 146 + job_id: j0pxze095 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -274,36 +274,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:18:58Z' + timestamp: '2024-08-27T00:33:27Z' - torchscript_onnx_tflite: - inference_time: 1148.0 - throughput: 871.0801393728223 + inference_time: 1079.0 + throughput: 926.7840593141798 estimated_peak_memory_range: - min: 24576 - max: 1652936 + min: 12288 + max: 1640512 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: jlpe6xd1g + total_layers: 145 + job_id: jlpen470p job_status: Passed torchscript_onnx_qnn: - inference_time: 1393.0 - throughput: 717.8750897343862 + inference_time: 1335.0 + throughput: 749.0636704119851 estimated_peak_memory_range: - min: 221184 - max: 1442256 + min: 229376 + max: 2017336 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: jo5m9kewg + total_layers: 146 + job_id: jo5mlv9qg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -312,36 +312,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:18:59Z' + timestamp: '2024-08-27T00:33:28Z' - torchscript_onnx_tflite: - inference_time: 3764.0 - throughput: 265.6748140276302 + inference_time: 4054.0 + throughput: 246.66995559940798 estimated_peak_memory_range: - min: 12288 - max: 49197216 + min: 77824 + max: 52408048 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: jygzzy3kg + total_layers: 145 + job_id: jygz0vl65 job_status: Passed torchscript_onnx_qnn: - inference_time: 5989.0 - throughput: 166.9727834362999 + inference_time: 5963.0 + throughput: 167.7008217340265 estimated_peak_memory_range: - min: 217088 - max: 8524448 + min: 258048 + max: 7847184 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: joprxd69p + total_layers: 146 + job_id: jopr71xeg job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -350,21 +350,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:19:01Z' + timestamp: '2024-08-27T00:33:30Z' - torchscript_onnx_tflite: - inference_time: 15587.0 - throughput: 64.15602745877975 + inference_time: 15444.0 + throughput: 64.75006475006475 estimated_peak_memory_range: - min: 45056 - max: 4011784 + min: 16384 + max: 2396200 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 146 + layers_on_npu: 145 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 146 - job_id: jz5wyzw6g + total_layers: 145 + job_id: jz5wrmljp job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -373,21 +373,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:18:50Z' + timestamp: '2024-08-27T00:33:21Z' - torchscript_onnx_qnn: - inference_time: 1592.0 - throughput: 628.1407035175879 + inference_time: 1462.0 + throughput: 683.9945280437756 estimated_peak_memory_range: - min: 389120 - max: 389120 + min: 430080 + max: 430080 primary_compute_unit: NPU precision: int8 layer_info: - layers_on_npu: 147 + layers_on_npu: 146 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 147 - job_id: jvgd64nep + total_layers: 146 + job_id: jvgdm7dlg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -396,4 +396,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:18:54Z' + timestamp: '2024-08-27T00:33:24Z' diff --git a/qai_hub_models/models/mnasnet05/perf.yaml b/qai_hub_models/models/mnasnet05/perf.yaml index 4e34e66b..dbe3d63c 100644 --- a/qai_hub_models/models/mnasnet05/perf.yaml +++ b/qai_hub_models/models/mnasnet05/perf.yaml @@ -45,11 +45,11 @@ models: - name: MNASNet05 performance_metrics: - torchscript_onnx_tflite: - inference_time: 766.0 - throughput: 1305.4830287206266 + inference_time: 753.0 + throughput: 1328.0212483399735 estimated_peak_memory_range: - min: 20480 - max: 1555464 + min: 24576 + max: 1629288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jlpe6xm0g + job_id: j1p3rq9zp job_status: Passed torchscript_onnx_qnn: - inference_time: 830.0 - throughput: 1204.8192771084337 + inference_time: 825.0 + throughput: 1212.121212121212 estimated_peak_memory_range: - min: 16384 - max: 169323560 + min: 28672 + max: 33938640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jmg9o2llg + job_id: jmg9q9mvp job_status: Passed torchscript_onnx: inference_time: 770.0 throughput: 1298.7012987012988 estimated_peak_memory_range: min: 36864 - max: 148524384 + max: 154293024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: joprxdv9p + job_id: jopr71leg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:17:52Z' + timestamp: '2024-08-27T00:32:25Z' - torchscript_onnx_tflite: - inference_time: 519.0 - throughput: 1926.7822736030828 + inference_time: 507.0 + throughput: 1972.3865877712033 estimated_peak_memory_range: - min: 12288 - max: 49519072 + min: 16384 + max: 50619312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jygzzyd6g + job_id: jwgo9erdg job_status: Passed torchscript_onnx_qnn: - inference_time: 562.0 - throughput: 1779.3594306049822 + inference_time: 556.0 + throughput: 1798.5611510791366 estimated_peak_memory_range: - min: 34201600 - max: 47266304 + min: 1900544 + max: 18112272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jnp1o1425 + job_id: jnp1mqjlp job_status: Passed torchscript_onnx: - inference_time: 553.0 - throughput: 1808.3182640144666 + inference_time: 551.0 + throughput: 1814.8820326678765 estimated_peak_memory_range: min: 0 - max: 54056320 + max: 54035152 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: jep2odk4g + job_id: jep2z3rmp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:17:53Z' + timestamp: '2024-08-27T00:32:26Z' - torchscript_onnx_tflite: - inference_time: 768.0 - throughput: 1302.0833333333333 + inference_time: 755.0 + throughput: 1324.5033112582782 estimated_peak_memory_range: min: 12288 - max: 151253928 + max: 8806712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jz5wyz6jg + job_id: j1pvnzdmg job_status: Passed torchscript_onnx_qnn: - inference_time: 789.0 - throughput: 1267.427122940431 + inference_time: 783.0 + throughput: 1277.139208173691 estimated_peak_memory_range: - min: 634880 - max: 2221048 + min: 638976 + max: 1875568 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jz57onylg + job_id: jz578verp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:17:46Z' + timestamp: '2024-08-27T00:32:20Z' - torchscript_onnx_tflite: - inference_time: 1072.0 - throughput: 932.8358208955224 + inference_time: 1027.0 + throughput: 973.7098344693281 estimated_peak_memory_range: - min: 16384 - max: 52136864 + min: 12288 + max: 51992272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jmg9o2nvg + job_id: j7gj8k785 job_status: Passed torchscript_onnx_qnn: - inference_time: 1116.0 - throughput: 896.0573476702509 + inference_time: 1091.0 + throughput: 916.5902841429881 estimated_peak_memory_range: - min: 618496 - max: 13813520 + min: 0 + max: 15162640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jegn1q6rp + job_id: jegnwrzmg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:17:51Z' + timestamp: '2024-08-27T00:32:24Z' - torchscript_onnx_tflite: - inference_time: 773.0 - throughput: 1293.6610608020699 + inference_time: 754.0 + throughput: 1326.2599469496022 estimated_peak_memory_range: - min: 57344 - max: 161677000 + min: 28672 + max: 162347600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jnp1o1zl5 + job_id: jlpen4z0p job_status: Passed torchscript_onnx_qnn: - inference_time: 793.0 - throughput: 1261.034047919294 + inference_time: 780.0 + throughput: 1282.051282051282 estimated_peak_memory_range: - min: 647168 - max: 1878752 + min: 634880 + max: 1994120 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jqp4e4lvg + job_id: jqp42jylg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:17:47Z' + timestamp: '2024-08-27T00:32:21Z' - torchscript_onnx_tflite: - inference_time: 787.0 - throughput: 1270.6480304955528 + inference_time: 754.0 + throughput: 1326.2599469496022 estimated_peak_memory_range: - min: 20480 - max: 1787896 + min: 45056 + max: 4157816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jvgd641lp + job_id: jygz0vm65 job_status: Passed torchscript_onnx_qnn: - inference_time: 789.0 - throughput: 1267.427122940431 + inference_time: 785.0 + throughput: 1273.8853503184714 estimated_peak_memory_range: - min: 20480 - max: 1704008 + min: 634880 + max: 2039832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: j0px0rk1p + job_id: j0pxzel95 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:17:48Z' + timestamp: '2024-08-27T00:32:22Z' - torchscript_onnx_tflite: - inference_time: 772.0 - throughput: 1295.3367875647668 + inference_time: 752.0 + throughput: 1329.787234042553 estimated_peak_memory_range: - min: 20480 - max: 5116400 + min: 12288 + max: 2984080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jz5wyze6g + job_id: jz5wrm7jp job_status: Passed torchscript_onnx_qnn: - inference_time: 795.0 - throughput: 1257.861635220126 + inference_time: 784.0 + throughput: 1275.5102040816328 estimated_peak_memory_range: - min: 634880 - max: 1853728 + min: 626688 + max: 1806160 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jo5m9knwg + job_id: jo5mlv0qg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:17:49Z' + timestamp: '2024-08-27T00:32:23Z' - torchscript_onnx_qnn: - inference_time: 934.0 - throughput: 1070.6638115631692 + inference_time: 1065.0 + throughput: 938.9671361502348 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jvgd64xep + job_id: jvgdm73lg job_status: Passed torchscript_onnx: - inference_time: 826.0 - throughput: 1210.6537530266344 + inference_time: 825.0 + throughput: 1212.121212121212 estimated_peak_memory_range: - min: 6332416 - max: 6332416 + min: 8376320 + max: 8376320 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 104 - job_id: jqpy8217g + job_id: jqpyyvo4p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:17:54Z' + timestamp: '2024-08-27T00:32:27Z' diff --git a/qai_hub_models/models/mobilenet_v2/perf.yaml b/qai_hub_models/models/mobilenet_v2/perf.yaml index a835506e..f13b4b7e 100644 --- a/qai_hub_models/models/mobilenet_v2/perf.yaml +++ b/qai_hub_models/models/mobilenet_v2/perf.yaml @@ -45,11 +45,11 @@ models: - name: MobileNet-v2 performance_metrics: - torchscript_onnx_tflite: - inference_time: 935.0 - throughput: 1069.51871657754 + inference_time: 902.0 + throughput: 1108.6474501108648 estimated_peak_memory_range: - min: 40960 - max: 1655504 + min: 16384 + max: 1547712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jz5wyynjg + job_id: j1p8k4d8p job_status: Passed torchscript_onnx_qnn: - inference_time: 1253.0 - throughput: 798.0845969672786 + inference_time: 1245.0 + throughput: 803.2128514056225 estimated_peak_memory_range: - min: 622592 - max: 41691128 + min: 618496 + max: 98625592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jo5m9kxqg + job_id: j1pvn9ymg job_status: Passed torchscript_onnx: - inference_time: 953.0 - throughput: 1049.3179433368311 + inference_time: 934.0 + throughput: 1070.6638115631692 estimated_peak_memory_range: min: 12288 - max: 9353248 + max: 9186832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jn5q41qmg + job_id: jz5wr1k6p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:17:13Z' + timestamp: '2024-08-27T11:54:11Z' - torchscript_onnx_tflite: - inference_time: 614.0 - throughput: 1628.6644951140065 + inference_time: 602.0 + throughput: 1661.1295681063123 estimated_peak_memory_range: - min: 12288 - max: 61931072 + min: 16384 + max: 62134576 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jmg9ooevg + job_id: jogkk9wog job_status: Passed torchscript_onnx_qnn: - inference_time: 841.0 - throughput: 1189.0606420927468 + inference_time: 835.0 + throughput: 1197.6047904191616 estimated_peak_memory_range: - min: 618496 - max: 16723456 + min: 0 + max: 14077248 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jegn1qvmp + job_id: j7gj8w685 job_status: Passed torchscript_onnx: - inference_time: 653.0 - throughput: 1531.3935681470139 + inference_time: 661.0 + throughput: 1512.8593040847202 estimated_peak_memory_range: min: 0 - max: 65098640 + max: 65303696 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: j1glw8mlp + job_id: jmg9qxrlp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:17:14Z' + timestamp: '2024-08-27T11:54:12Z' - torchscript_onnx_tflite: - inference_time: 918.0 - throughput: 1089.3246187363834 + inference_time: 901.0 + throughput: 1109.8779134295228 estimated_peak_memory_range: - min: 24576 - max: 1448864 + min: 12288 + max: 3578704 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jnp1ooxl5 + job_id: jn5qdmxmg job_status: Passed torchscript_onnx_qnn: - inference_time: 1184.0 - throughput: 844.5945945945946 + inference_time: 1190.0 + throughput: 840.3361344537815 estimated_peak_memory_range: min: 634880 - max: 1832056 + max: 2420752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jep2odymg + job_id: jygz04q65 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:17:07Z' + timestamp: '2024-08-27T11:54:07Z' - torchscript_onnx_tflite: - inference_time: 1122.0 - throughput: 891.2655971479501 + inference_time: 1092.0 + throughput: 915.7509157509157 estimated_peak_memory_range: min: 16384 - max: 64834112 + max: 63762416 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jvgd66llp + job_id: j1glq19lp job_status: Passed torchscript_onnx_qnn: - inference_time: 1454.0 - throughput: 687.757909215956 + inference_time: 1429.0 + throughput: 699.7900629811056 estimated_peak_memory_range: min: 618496 - max: 19515984 + max: 18375392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jogk60xo5 + job_id: jvgdmzylg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:17:12Z' + timestamp: '2024-08-27T11:54:10Z' - torchscript_onnx_tflite: - inference_time: 918.0 - throughput: 1089.3246187363834 + inference_time: 899.0 + throughput: 1112.3470522803113 estimated_peak_memory_range: - min: 32768 - max: 1450864 + min: 40960 + max: 166292984 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jz57onrrg + job_id: jw560d975 job_status: Passed torchscript_onnx_qnn: - inference_time: 1188.0 - throughput: 841.7508417508418 + inference_time: 1192.0 + throughput: 838.9261744966443 estimated_peak_memory_range: - min: 634880 - max: 1946232 + min: 626688 + max: 1930728 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jqpy8234g + job_id: jz5wr10jp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:17:08Z' + timestamp: '2024-08-27T11:54:07Z' - torchscript_onnx_tflite: - inference_time: 929.0 - throughput: 1076.4262648008612 + inference_time: 899.0 + throughput: 1112.3470522803113 estimated_peak_memory_range: - min: 40960 - max: 1493880 + min: 28672 + max: 2128080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jqp4e4rlg + job_id: j1p3rwlzp job_status: Passed torchscript_onnx_qnn: - inference_time: 1194.0 - throughput: 837.5209380234506 + inference_time: 1192.0 + throughput: 838.9261744966443 estimated_peak_memory_range: - min: 638976 - max: 2164240 + min: 634880 + max: 2114328 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: j2p0o90ep + job_id: jmg9qx7vp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:17:09Z' + timestamp: '2024-08-27T11:54:08Z' - torchscript_onnx_tflite: - inference_time: 936.0 - throughput: 1068.3760683760684 + inference_time: 902.0 + throughput: 1108.6474501108648 estimated_peak_memory_range: - min: 24576 - max: 1509648 + min: 40960 + max: 53198736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: j0px0ro9p + job_id: jwgo947dg job_status: Passed torchscript_onnx_qnn: - inference_time: 1209.0 - throughput: 827.129859387924 + inference_time: 1199.0 + throughput: 834.0283569641368 estimated_peak_memory_range: - min: 618496 - max: 1905280 + min: 626688 + max: 1955896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: j1p8jry85 + job_id: jnp1mvklp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:17:11Z' + timestamp: '2024-08-27T11:54:09Z' - torchscript_onnx_qnn: - inference_time: 1365.0 - throughput: 732.6007326007326 + inference_time: 1477.0 + throughput: 677.0480704129993 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: joprxd3ep + job_id: jlpenl00p job_status: Passed torchscript_onnx: - inference_time: 956.0 - throughput: 1046.0251046025105 + inference_time: 984.0 + throughput: 1016.260162601626 estimated_peak_memory_range: - min: 9072640 - max: 9072640 + min: 10612736 + max: 10612736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 105 - job_id: jw56om475 + job_id: jnp1mv92p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:17:15Z' + timestamp: '2024-08-27T11:54:13Z' diff --git a/qai_hub_models/models/mobilenet_v2_quantized/perf.yaml b/qai_hub_models/models/mobilenet_v2_quantized/perf.yaml index 5f3f9e98..92940dbd 100644 --- a/qai_hub_models/models/mobilenet_v2_quantized/perf.yaml +++ b/qai_hub_models/models/mobilenet_v2_quantized/perf.yaml @@ -51,64 +51,11 @@ models: - name: MobileNet-v2-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 287.0 - throughput: 3484.320557491289 - estimated_peak_memory_range: - min: 36864 - max: 102311264 - primary_compute_unit: NPU - precision: int8 - layer_info: - layers_on_npu: 74 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 74 - job_id: jygzzz96g - job_status: Passed - torchscript_onnx_qnn: - inference_time: 669.0 - throughput: 1494.7683109118086 - estimated_peak_memory_range: - min: 16384 - max: 15703096 - primary_compute_unit: NPU - precision: int8 - layer_info: - layers_on_npu: 71 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 71 - job_id: jegn118mp - job_status: Passed - torchscript_onnx: - inference_time: 534.0 - throughput: 1872.6591760299625 - estimated_peak_memory_range: - min: 12288 - max: 46077256 - primary_compute_unit: NPU - precision: int8 - layer_info: - layers_on_npu: 74 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 74 - job_id: jw56ool75 - job_status: Passed - reference_device_info: - name: Samsung Galaxy S23 - os: '13' - form_factor: Phone - os_name: Android - manufacturer: Samsung - chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:16:34Z' - - torchscript_onnx_tflite: - inference_time: 209.0 - throughput: 4784.688995215311 + inference_time: 198.0 + throughput: 5050.50505050505 estimated_peak_memory_range: min: 12288 - max: 40947312 + max: 42229808 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: jz5wyyvjg + job_id: jz5wrm1zp job_status: Passed torchscript_onnx_qnn: inference_time: 475.0 throughput: 2105.2631578947367 estimated_peak_memory_range: - min: 12288 - max: 16169216 + min: 0 + max: 17734192 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: joprxxjep + job_id: jqp42j3lg job_status: Passed torchscript_onnx: - inference_time: 406.0 - throughput: 2463.054187192118 + inference_time: 390.0 + throughput: 2564.102564102564 estimated_peak_memory_range: - min: 0 - max: 60453424 + min: 12288 + max: 61227152 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: j1p3oo2zp + job_id: jogkkr8og job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:16:35Z' + timestamp: '2024-08-27T00:31:16Z' - torchscript_onnx_tflite: - inference_time: 293.0 - throughput: 3412.9692832764504 + inference_time: 285.0 + throughput: 3508.7719298245615 estimated_peak_memory_range: min: 12288 - max: 9532080 + max: 1317448 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: jmg9oo1vg + job_id: jmg9q9xqp job_status: Passed torchscript_onnx_qnn: - inference_time: 604.0 - throughput: 1655.6291390728477 + inference_time: 608.0 + throughput: 1644.7368421052631 estimated_peak_memory_range: - min: 180224 - max: 1358064 + min: 184320 + max: 1751600 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +131,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jqpy8804g + job_id: jo5mlv8qg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +140,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:16:27Z' + timestamp: '2024-08-27T00:31:09Z' - torchscript_onnx_tflite: - inference_time: 351.0 - throughput: 2849.002849002849 + inference_time: 330.0 + throughput: 3030.3030303030305 estimated_peak_memory_range: - min: 12288 - max: 41054736 + min: 16384 + max: 41569120 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +154,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: jnp1ooll5 + job_id: jnp1mqvkp job_status: Passed torchscript_onnx_qnn: - inference_time: 721.0 - throughput: 1386.9625520110958 + inference_time: 718.0 + throughput: 1392.757660167131 estimated_peak_memory_range: min: 159744 - max: 15366160 + max: 18161920 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +169,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jn5q44omg + job_id: jqpyyvm4p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +178,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:16:32Z' + timestamp: '2024-08-27T00:31:13Z' - torchscript_onnx_tflite: - inference_time: 315.0 - throughput: 3174.6031746031745 + inference_time: 280.0 + throughput: 3571.4285714285716 estimated_peak_memory_range: min: 12288 - max: 6689632 + max: 103488840 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +192,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: jvgd669lp + job_id: jvgdm7zkg job_status: Passed torchscript_onnx_qnn: - inference_time: 611.0 - throughput: 1636.6612111292961 + inference_time: 606.0 + throughput: 1650.1650165016501 estimated_peak_memory_range: - min: 180224 - max: 1419800 + min: 176128 + max: 1813936 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +207,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: j2p0oo7ep + job_id: jegnwrkmg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +216,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:16:28Z' + timestamp: '2024-08-27T00:31:10Z' - torchscript_onnx_tflite: - inference_time: 293.0 - throughput: 3412.9692832764504 + inference_time: 279.0 + throughput: 3584.2293906810037 estimated_peak_memory_range: - min: 12288 - max: 1322488 + min: 24576 + max: 2929472 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +230,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: jz57oo3rg + job_id: jz5wrm1jp job_status: Passed torchscript_onnx_qnn: - inference_time: 607.0 - throughput: 1647.4464579901153 + inference_time: 604.0 + throughput: 1655.6291390728477 estimated_peak_memory_range: - min: 172032 - max: 1377024 + min: 176128 + max: 1530792 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +245,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: j1p8jjv85 + job_id: jopr71weg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +254,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:16:30Z' + timestamp: '2024-08-27T00:31:11Z' - torchscript_onnx_tflite: - inference_time: 285.0 - throughput: 3508.7719298245615 + inference_time: 280.0 + throughput: 3571.4285714285716 estimated_peak_memory_range: - min: 12288 - max: 1381336 + min: 36864 + max: 1467088 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +268,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: jqp4ee0lg + job_id: jmg9q9xvp job_status: Passed torchscript_onnx_qnn: - inference_time: 601.0 - throughput: 1663.8935108153078 + inference_time: 602.0 + throughput: 1661.1295681063123 estimated_peak_memory_range: - min: 180224 - max: 1355016 + min: 212992 + max: 1794656 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +283,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jogk66mo5 + job_id: jep2z3emp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +292,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:16:31Z' + timestamp: '2024-08-27T00:31:12Z' - torchscript_onnx_tflite: - inference_time: 828.0 - throughput: 1207.729468599034 + inference_time: 800.0 + throughput: 1250.0 estimated_peak_memory_range: min: 12288 - max: 26521968 + max: 27091792 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +306,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: j0px0029p + job_id: jnp1mqvlp job_status: Passed torchscript_onnx_qnn: - inference_time: 1425.0 - throughput: 701.7543859649123 + inference_time: 1472.0 + throughput: 679.3478260869565 estimated_peak_memory_range: - min: 45056 - max: 8239744 + min: 12288 + max: 8440896 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +321,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: j1glwwrlp + job_id: j2p0xe6ep job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +330,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:16:33Z' + timestamp: '2024-08-27T00:31:14Z' - torchscript_onnx_tflite: - inference_time: 7747.0 - throughput: 129.08222537756552 + inference_time: 7527.0 + throughput: 132.85505513484787 estimated_peak_memory_range: - min: 274432 - max: 6787704 + min: 32768 + max: 11225400 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +344,7 @@ models: layers_on_gpu: 2 layers_on_cpu: 0 total_layers: 74 - job_id: jo5m99yqg + job_id: jvgdm7zlg job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +353,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:16:23Z' + timestamp: '2024-08-27T00:31:05Z' - torchscript_onnx_qnn: - inference_time: 744.0 - throughput: 1344.0860215053763 + inference_time: 768.0 + throughput: 1302.0833333333333 estimated_peak_memory_range: - min: 569344 - max: 569344 + min: 548864 + max: 548864 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +367,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jep2oonmg + job_id: j0pxzex95 job_status: Passed torchscript_onnx: - inference_time: 589.0 - throughput: 1697.792869269949 + inference_time: 556.0 + throughput: 1798.5611510791366 estimated_peak_memory_range: - min: 6971392 - max: 6971392 + min: 6414336 + max: 6414336 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +382,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: jwgoddqd5 + job_id: jn5qd9vmg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +391,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:16:36Z' + timestamp: '2024-08-27T00:31:17Z' diff --git a/qai_hub_models/models/mobilenet_v3_large/perf.yaml b/qai_hub_models/models/mobilenet_v3_large/perf.yaml index 7e199d98..8bd0ebb4 100644 --- a/qai_hub_models/models/mobilenet_v3_large/perf.yaml +++ b/qai_hub_models/models/mobilenet_v3_large/perf.yaml @@ -45,11 +45,11 @@ models: - name: MobileNet-v3-Large performance_metrics: - torchscript_onnx_tflite: - inference_time: 1026.0 - throughput: 974.6588693957115 + inference_time: 986.0 + throughput: 1014.1987829614604 estimated_peak_memory_range: - min: 57344 - max: 1655496 + min: 20480 + max: 1554032 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jmg9oovvg + job_id: jmg9q9kqp job_status: Passed torchscript_onnx_qnn: - inference_time: 1049.0 - throughput: 953.2888465204957 + inference_time: 1039.0 + throughput: 962.4639076034649 estimated_peak_memory_range: - min: 20480 - max: 67954576 + min: 2129920 + max: 69544424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 144 - job_id: jegn11ymp + job_id: jegnwr7vg job_status: Passed torchscript_onnx: - inference_time: 1036.0 - throughput: 965.2509652509652 + inference_time: 1026.0 + throughput: 974.6588693957115 estimated_peak_memory_range: - min: 32768 - max: 65718720 + min: 12288 + max: 16027544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 162 - job_id: j1glwwzlp + job_id: j1glqe1ep job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:15:43Z' + timestamp: '2024-08-27T00:30:28Z' - torchscript_onnx_tflite: - inference_time: 683.0 - throughput: 1464.1288433382138 + inference_time: 672.0 + throughput: 1488.095238095238 estimated_peak_memory_range: min: 16384 - max: 66437904 + max: 66408016 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jnp1oo0l5 + job_id: jnp1mq7kp job_status: Passed torchscript_onnx_qnn: - inference_time: 716.0 - throughput: 1396.6480446927374 + inference_time: 710.0 + throughput: 1408.4507042253522 estimated_peak_memory_range: min: 618496 - max: 20352640 + max: 18222416 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 144 - job_id: joprxxqep + job_id: jopr71nvg job_status: Passed torchscript_onnx: - inference_time: 728.0 - throughput: 1373.6263736263736 + inference_time: 741.0 + throughput: 1349.527665317139 estimated_peak_memory_range: min: 0 - max: 99897840 + max: 100374256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 162 - job_id: jw56ooj75 + job_id: jw560qdv5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:15:44Z' + timestamp: '2024-08-27T00:30:29Z' - torchscript_onnx_tflite: - inference_time: 1011.0 - throughput: 989.1196834817013 + inference_time: 983.0 + throughput: 1017.293997965412 estimated_peak_memory_range: - min: 57344 - max: 4768720 + min: 36864 + max: 232856552 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jvgd66wlp + job_id: jvgdm78kg job_status: Passed torchscript_onnx_qnn: - inference_time: 993.0 - throughput: 1007.0493454179255 + inference_time: 994.0 + throughput: 1006.0362173038229 estimated_peak_memory_range: - min: 630784 - max: 1938888 + min: 634880 + max: 1929240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 144 - job_id: jqpy88w4g + job_id: jqpyyv7rp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:15:37Z' + timestamp: '2024-08-27T00:30:23Z' - torchscript_onnx_tflite: - inference_time: 1416.0 - throughput: 706.2146892655368 + inference_time: 1374.0 + throughput: 727.802037845706 estimated_peak_memory_range: min: 16384 - max: 68832048 + max: 68161248 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jz57oowrg + job_id: jz578vkqp job_status: Passed torchscript_onnx_qnn: - inference_time: 1468.0 - throughput: 681.1989100817439 + inference_time: 1454.0 + throughput: 687.757909215956 estimated_peak_memory_range: - min: 618496 - max: 20414560 + min: 622592 + max: 20430080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jn5q44kmg + job_id: jn5qd9m7g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:15:41Z' + timestamp: '2024-08-27T00:30:27Z' - torchscript_onnx_tflite: - inference_time: 1021.0 - throughput: 979.4319294809011 + inference_time: 985.0 + throughput: 1015.2284263959391 estimated_peak_memory_range: - min: 12288 - max: 1505232 + min: 73728 + max: 250947600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jqp4eeolg + job_id: jqp42jmqg job_status: Passed torchscript_onnx_qnn: - inference_time: 999.0 - throughput: 1001.001001001001 + inference_time: 1002.0 + throughput: 998.003992015968 estimated_peak_memory_range: min: 634880 - max: 2033304 + max: 2110688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 144 - job_id: j2p0ooqep + job_id: j2p0xev2p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:15:38Z' + timestamp: '2024-08-27T00:30:24Z' - torchscript_onnx_tflite: - inference_time: 1031.0 - throughput: 969.9321047526673 + inference_time: 988.0 + throughput: 1012.1457489878543 estimated_peak_memory_range: min: 12288 - max: 8333816 + max: 2089664 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: j0px00j9p + job_id: j0pxze3j5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1004.0 - throughput: 996.01593625498 + inference_time: 1005.0 + throughput: 995.0248756218906 estimated_peak_memory_range: - min: 647168 - max: 2112248 + min: 634880 + max: 2299504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 144 - job_id: j1p8jj985 + job_id: j1p8kw4zp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:15:39Z' + timestamp: '2024-08-27T00:30:25Z' - torchscript_onnx_tflite: - inference_time: 1023.0 - throughput: 977.5171065493646 + inference_time: 986.0 + throughput: 1014.1987829614604 estimated_peak_memory_range: - min: 28672 - max: 3536920 + min: 57344 + max: 240464320 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jo5m992qg + job_id: jo5mlvoyg job_status: Passed torchscript_onnx_qnn: - inference_time: 1002.0 - throughput: 998.003992015968 + inference_time: 991.0 + throughput: 1009.0817356205853 estimated_peak_memory_range: - min: 638976 - max: 1907936 + min: 634880 + max: 1763096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 144 - job_id: jogk66no5 + job_id: jogkkr9yg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:15:40Z' + timestamp: '2024-08-27T00:30:26Z' - torchscript_onnx_qnn: - inference_time: 1246.0 - throughput: 802.5682182985554 + inference_time: 1254.0 + throughput: 797.4481658692185 estimated_peak_memory_range: - min: 602112 - max: 602112 + min: 1191936 + max: 1191936 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 144 - job_id: jep2oo6mg + job_id: jep2z3vxp job_status: Passed torchscript_onnx: - inference_time: 1061.0 - throughput: 942.5070688030161 + inference_time: 1105.0 + throughput: 904.9773755656108 estimated_peak_memory_range: - min: 14831616 - max: 14831616 + min: 14811136 + max: 14811136 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 162 - job_id: j1p3oo3zp + job_id: j1p3rqwxp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:15:45Z' + timestamp: '2024-08-27T00:30:30Z' diff --git a/qai_hub_models/models/mobilenet_v3_large_quantized/perf.yaml b/qai_hub_models/models/mobilenet_v3_large_quantized/perf.yaml index 90075a01..e1dc4ed8 100644 --- a/qai_hub_models/models/mobilenet_v3_large_quantized/perf.yaml +++ b/qai_hub_models/models/mobilenet_v3_large_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: MobileNet-v3-Large-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 362.0 - throughput: 2762.4309392265195 + inference_time: 343.0 + throughput: 2915.451895043732 estimated_peak_memory_range: - min: 40960 - max: 1579944 + min: 12288 + max: 2612264 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: jz5wyyjjg + job_id: j2p0xvxep job_status: Passed torchscript_onnx_qnn: - inference_time: 625.0 - throughput: 1600.0 + inference_time: 613.0 + throughput: 1631.3213703099511 estimated_peak_memory_range: - min: 20480 - max: 14588016 + min: 28672 + max: 15105752 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: joprxxkep + job_id: j7gj8w885 job_status: Passed torchscript_onnx: - inference_time: 5353.0 - throughput: 186.81113394358303 + inference_time: 5384.0 + throughput: 185.73551263001485 estimated_peak_memory_range: - min: 25489408 - max: 27250488 + min: 23101440 + max: 24822544 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 24 total_layers: 171 - job_id: j1p3ookzp + job_id: j0pxzd895 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:15:02Z' + timestamp: '2024-08-27T11:53:38Z' - torchscript_onnx_tflite: - inference_time: 249.0 - throughput: 4016.0642570281125 + inference_time: 245.0 + throughput: 4081.6326530612246 estimated_peak_memory_range: - min: 12288 - max: 52414544 + min: 16384 + max: 52657376 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: jmg9oo6vg + job_id: j1p8k4k8p job_status: Passed torchscript_onnx_qnn: - inference_time: 456.0 - throughput: 2192.9824561403507 + inference_time: 455.0 + throughput: 2197.802197802198 estimated_peak_memory_range: - min: 0 - max: 15659728 + min: 163840 + max: 16288000 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jep2oo8mg + job_id: jlpenln0p job_status: Passed torchscript_onnx: - inference_time: 4434.0 - throughput: 225.52999548940008 + inference_time: 4206.0 + throughput: 237.75558725630052 estimated_peak_memory_range: - min: 23580672 - max: 100917536 + min: 24014848 + max: 102104688 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 24 total_layers: 171 - job_id: jwgoddyd5 + job_id: jo5mld1qg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:15:03Z' + timestamp: '2024-08-27T11:53:39Z' - torchscript_onnx_tflite: - inference_time: 370.0 - throughput: 2702.7027027027025 + inference_time: 342.0 + throughput: 2923.9766081871344 estimated_peak_memory_range: min: 12288 - max: 1541216 + max: 1428648 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: jnp1oorl5 + job_id: jogkk9kog job_status: Passed torchscript_onnx_qnn: inference_time: 572.0 throughput: 1748.2517482517483 estimated_peak_memory_range: - min: 176128 - max: 1448752 + min: 184320 + max: 1590112 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j2p0ooyep + job_id: jz5wr1rjp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:14:55Z' + timestamp: '2024-08-27T11:53:32Z' - torchscript_onnx_tflite: - inference_time: 455.0 - throughput: 2197.802197802198 + inference_time: 445.0 + throughput: 2247.191011235955 estimated_peak_memory_range: - min: 28672 - max: 52788880 + min: 16384 + max: 53808448 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: jvgd66jlp + job_id: jn5qdmdmg job_status: Passed torchscript_onnx_qnn: - inference_time: 780.0 - throughput: 1282.051282051282 + inference_time: 755.0 + throughput: 1324.5033112582782 estimated_peak_memory_range: min: 163840 - max: 20096560 + max: 17155216 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1glwwnlp + job_id: jz57871rp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:15:00Z' + timestamp: '2024-08-27T11:53:36Z' - torchscript_onnx_tflite: - inference_time: 364.0 - throughput: 2747.252747252747 + inference_time: 340.0 + throughput: 2941.176470588235 estimated_peak_memory_range: - min: 12288 - max: 126105048 + min: 28672 + max: 117101296 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: jz57oozrg + job_id: j1glq1qlp job_status: Passed torchscript_onnx_qnn: inference_time: 572.0 throughput: 1748.2517482517483 estimated_peak_memory_range: min: 184320 - max: 1757400 + max: 1739560 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1p8jjo85 + job_id: jmg9qxqvp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:14:56Z' + timestamp: '2024-08-27T11:53:33Z' - torchscript_onnx_tflite: - inference_time: 363.0 - throughput: 2754.8209366391184 + inference_time: 342.0 + throughput: 2923.9766081871344 estimated_peak_memory_range: - min: 16384 - max: 1375832 + min: 12288 + max: 26212928 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: jqp4eeqlg + job_id: jw560d075 job_status: Passed torchscript_onnx_qnn: - inference_time: 565.0 - throughput: 1769.9115044247787 + inference_time: 573.0 + throughput: 1745.2006980802792 estimated_peak_memory_range: min: 184320 - max: 1778952 + max: 1452136 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jogk66zo5 + job_id: jnp1mvmlp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:14:58Z' + timestamp: '2024-08-27T11:53:34Z' - torchscript_onnx_tflite: - inference_time: 350.0 - throughput: 2857.1428571428573 + inference_time: 342.0 + throughput: 2923.9766081871344 estimated_peak_memory_range: min: 12288 - max: 7353848 + max: 14925536 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: j0px00v9p + job_id: j1p3rwrzp job_status: Passed torchscript_onnx_qnn: - inference_time: 563.0 - throughput: 1776.1989342806394 + inference_time: 560.0 + throughput: 1785.7142857142858 estimated_peak_memory_range: min: 184320 - max: 1412384 + max: 1513096 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jn5q448mg + job_id: jvgdmzmlg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:14:59Z' + timestamp: '2024-08-27T11:53:35Z' - torchscript_onnx_tflite: - inference_time: 1178.0 - throughput: 848.8964346349745 + inference_time: 1115.0 + throughput: 896.8609865470852 estimated_peak_memory_range: min: 12288 - max: 32538240 + max: 33603568 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: jo5m99rqg + job_id: jwgo949dg job_status: Passed torchscript_onnx_qnn: - inference_time: 1676.0 - throughput: 596.6587112171837 + inference_time: 1767.0 + throughput: 565.9309564233164 estimated_peak_memory_range: - min: 12288 - max: 8512496 + min: 172032 + max: 8067872 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jw56oo675 + job_id: jqp4296lg job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:15:01Z' + timestamp: '2024-08-27T11:53:37Z' - torchscript_onnx_tflite: - inference_time: 6817.0 - throughput: 146.69209329617135 + inference_time: 6833.0 + throughput: 146.34860237084735 estimated_peak_memory_range: - min: 16384 - max: 8042640 + min: 12288 + max: 4412720 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 137 - job_id: jegn112mp + job_id: j1pvn9nmg job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:14:50Z' + timestamp: '2024-08-27T11:53:28Z' - torchscript_onnx_qnn: - inference_time: 707.0 - throughput: 1414.4271570014143 + inference_time: 694.0 + throughput: 1440.922190201729 estimated_peak_memory_range: - min: 524288 - max: 524288 + min: 458752 + max: 458752 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jqpy88e4g + job_id: jygz04065 job_status: Passed torchscript_onnx: - inference_time: 4787.0 - throughput: 208.89910173386255 + inference_time: 4825.0 + throughput: 207.2538860103627 estimated_peak_memory_range: - min: 35037184 - max: 35037184 + min: 35123200 + max: 35123200 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 24 total_layers: 171 - job_id: j1pv223mg + job_id: jegnw7dmg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:15:05Z' + timestamp: '2024-08-27T11:53:39Z' diff --git a/qai_hub_models/models/mobilenet_v3_small/perf.yaml b/qai_hub_models/models/mobilenet_v3_small/perf.yaml index 5bcffb6c..937ba778 100644 --- a/qai_hub_models/models/mobilenet_v3_small/perf.yaml +++ b/qai_hub_models/models/mobilenet_v3_small/perf.yaml @@ -45,11 +45,11 @@ models: - name: MobileNet-v3-Small performance_metrics: - torchscript_onnx_tflite: - inference_time: 844.0 - throughput: 1184.8341232227488 + inference_time: 816.0 + throughput: 1225.4901960784314 estimated_peak_memory_range: min: 12288 - max: 1432592 + max: 1365784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 115 - job_id: j0px00qjp + job_id: jnp1mqqkp job_status: Passed torchscript_onnx_qnn: - inference_time: 882.0 - throughput: 1133.7868480725624 + inference_time: 870.0 + throughput: 1149.4252873563219 estimated_peak_memory_range: - min: 12288 - max: 144938808 + min: 16384 + max: 183799992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1p8jjlz5 + job_id: jopr719vg job_status: Passed torchscript_onnx: - inference_time: 831.0 - throughput: 1203.3694344163657 + inference_time: 827.0 + throughput: 1209.1898428053205 estimated_peak_memory_range: - min: 12288 - max: 7216032 + min: 36864 + max: 7322872 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: j7gj33j7p + job_id: jw560qvv5 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:14:08Z' + timestamp: '2024-08-27T00:29:03Z' - torchscript_onnx_tflite: - inference_time: 548.0 - throughput: 1824.8175182481752 + inference_time: 538.0 + throughput: 1858.736059479554 estimated_peak_memory_range: - min: 24576 - max: 44836896 + min: 12288 + max: 45268480 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 115 - job_id: jo5m997yg + job_id: jvgdm77kg job_status: Passed torchscript_onnx_qnn: - inference_time: 574.0 - throughput: 1742.1602787456445 + inference_time: 573.0 + throughput: 1745.2006980802792 estimated_peak_memory_range: - min: 339968 - max: 13260112 + min: 20480 + max: 14898400 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jogk66jy5 + job_id: jep2z3jxp job_status: Passed torchscript_onnx: - inference_time: 591.0 - throughput: 1692.047377326565 + inference_time: 587.0 + throughput: 1703.5775127768313 estimated_peak_memory_range: min: 0 - max: 59042896 + max: 59804368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jlpe66j7g + job_id: j1p3rq8xp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:14:09Z' + timestamp: '2024-08-27T00:29:04Z' - torchscript_onnx_tflite: - inference_time: 844.0 - throughput: 1184.8341232227488 + inference_time: 811.0 + throughput: 1233.0456226880394 estimated_peak_memory_range: - min: 61440 - max: 1589376 + min: 32768 + max: 1567264 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 115 - job_id: jegn11jvp + job_id: jz578vvqp job_status: Passed torchscript_onnx_qnn: - inference_time: 857.0 - throughput: 1166.8611435239206 + inference_time: 849.0 + throughput: 1177.8563015312131 estimated_peak_memory_range: - min: 634880 - max: 2166808 + min: 630784 + max: 2262000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1glwwjep + job_id: j2p0xek2p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:14:02Z' + timestamp: '2024-08-27T00:28:58Z' - torchscript_onnx_tflite: - inference_time: 1140.0 - throughput: 877.1929824561404 + inference_time: 1095.0 + throughput: 913.2420091324201 estimated_peak_memory_range: min: 16384 - max: 46938128 + max: 46679488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 115 - job_id: joprxxzvp + job_id: jqp42jjqg job_status: Passed torchscript_onnx_qnn: - inference_time: 1161.0 - throughput: 861.3264427217915 + inference_time: 1160.0 + throughput: 862.0689655172414 estimated_peak_memory_range: - min: 1060864 - max: 17652544 + min: 618496 + max: 17398128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: j1pv22j7g + job_id: j1glqe7ep job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:14:07Z' + timestamp: '2024-08-27T00:29:02Z' - torchscript_onnx_tflite: - inference_time: 840.0 - throughput: 1190.4761904761904 + inference_time: 816.0 + throughput: 1225.4901960784314 estimated_peak_memory_range: - min: 32768 - max: 1469376 + min: 20480 + max: 1544072 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 115 - job_id: jep2oo2xg + job_id: j0pxzeej5 job_status: Passed torchscript_onnx_qnn: - inference_time: 859.0 - throughput: 1164.1443538998835 + inference_time: 847.0 + throughput: 1180.637544273908 estimated_peak_memory_range: - min: 28672 - max: 1276072 + min: 634880 + max: 2197048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jw56ookv5 + job_id: j1p8kw8zp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:14:03Z' + timestamp: '2024-08-27T00:28:59Z' - torchscript_onnx_tflite: - inference_time: 832.0 - throughput: 1201.923076923077 + inference_time: 815.0 + throughput: 1226.993865030675 estimated_peak_memory_range: - min: 12288 - max: 1412400 + min: 40960 + max: 1493568 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 115 - job_id: jqpy889rg + job_id: jo5mlvvyg job_status: Passed torchscript_onnx_qnn: - inference_time: 856.0 - throughput: 1168.2242990654206 + inference_time: 846.0 + throughput: 1182.033096926714 estimated_peak_memory_range: - min: 634880 - max: 2117528 + min: 630784 + max: 1931376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1p3ooyxp + job_id: jogkkrdyg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:14:04Z' + timestamp: '2024-08-27T00:29:00Z' - torchscript_onnx_tflite: - inference_time: 839.0 - throughput: 1191.8951132300358 + inference_time: 815.0 + throughput: 1226.993865030675 estimated_peak_memory_range: - min: 32768 - max: 45983784 + min: 28672 + max: 160005336 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 115 - job_id: j2p0oon2p + job_id: jegnwrxvg job_status: Passed torchscript_onnx_qnn: - inference_time: 863.0 - throughput: 1158.7485515643104 + inference_time: 850.0 + throughput: 1176.4705882352941 estimated_peak_memory_range: - min: 651264 - max: 1948104 + min: 634880 + max: 2131472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jwgoddj45 + job_id: jn5qd9w7g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:14:06Z' + timestamp: '2024-08-27T00:29:01Z' - torchscript_onnx_qnn: - inference_time: 1129.0 - throughput: 885.7395925597874 + inference_time: 1043.0 + throughput: 958.7727708533077 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jn5q44j7g + job_id: jqpyyvnrp job_status: Passed torchscript_onnx: - inference_time: 860.0 - throughput: 1162.7906976744187 + inference_time: 936.0 + throughput: 1068.3760683760684 estimated_peak_memory_range: - min: 7561216 - max: 7561216 + min: 8646656 + max: 8646656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jygzzz1zg + job_id: jwgo9em4g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:14:10Z' + timestamp: '2024-08-27T00:29:05Z' diff --git a/qai_hub_models/models/openai_clip/export.py b/qai_hub_models/models/openai_clip/export.py index f60d3992..0bb149d7 100644 --- a/qai_hub_models/models/openai_clip/export.py +++ b/qai_hub_models/models/openai_clip/export.py @@ -128,8 +128,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/openai_clip/perf.yaml b/qai_hub_models/models/openai_clip/perf.yaml index 74ada990..5ce62a80 100644 --- a/qai_hub_models/models/openai_clip/perf.yaml +++ b/qai_hub_models/models/openai_clip/perf.yaml @@ -45,26 +45,26 @@ models: - name: CLIPTextEncoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 11739.0 - throughput: 85.1861316977596 + inference_time: 5809.0 + throughput: 172.14666896195558 estimated_peak_memory_range: - min: 16384 - max: 8834184 + min: 61440 + max: 2471208 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 573 - layers_on_gpu: 1 + layers_on_npu: 574 + layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 576 - job_id: jep2ooqxg + job_id: jn5qdm4mg job_status: Passed torchscript_onnx_qnn: - inference_time: 7795.0 - throughput: 128.2873636946761 + inference_time: 6029.0 + throughput: 165.8649859014762 estimated_peak_memory_range: - min: 12288 - max: 17129224 + min: 36864 + max: 23137784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 377 - job_id: jz5wyy0zg + job_id: j0pxzdr95 job_status: Passed torchscript_onnx: - inference_time: 31404.0 - throughput: 31.84307731499172 + inference_time: 30402.0 + throughput: 32.892572857048876 estimated_peak_memory_range: - min: 90112 - max: 136922016 + min: 65536 + max: 137179496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 389 - job_id: j1glwwdep + job_id: jlpenlx0p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,28 +96,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:13:20Z' + timestamp: '2024-08-27T11:52:44Z' - torchscript_onnx_tflite: - inference_time: 8182.0 - throughput: 122.21950623319482 + inference_time: 4028.0 + throughput: 248.26216484607747 estimated_peak_memory_range: min: 16384 - max: 221756144 + max: 169204080 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 573 - layers_on_gpu: 1 + layers_on_npu: 574 + layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 576 - job_id: j2p0oo82p + job_id: jw560do75 job_status: Passed torchscript_onnx_qnn: - inference_time: 5479.0 - throughput: 182.5150574922431 + inference_time: 4174.0 + throughput: 239.57834211787255 estimated_peak_memory_range: - min: 12288 - max: 57987488 + min: 823296 + max: 71011280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 377 - job_id: jnp1ookk5 + job_id: jegnw7qmg job_status: Passed torchscript_onnx: - inference_time: 22198.0 - throughput: 45.04910352283989 + inference_time: 21481.0 + throughput: 46.55276756203156 estimated_peak_memory_range: - min: 49152 - max: 437223600 + min: 57344 + max: 441770048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 389 - job_id: j1p3oodxp + job_id: jz5wr1zjp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,28 +149,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:13:22Z' + timestamp: '2024-08-27T11:52:46Z' - torchscript_onnx_tflite: - inference_time: 11609.0 - throughput: 86.14006374364718 + inference_time: 5766.0 + throughput: 173.4304543877905 estimated_peak_memory_range: min: 24576 - max: 8377648 + max: 2416080 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 573 - layers_on_gpu: 1 + layers_on_npu: 574 + layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 576 - job_id: jogk66wy5 + job_id: jwgo94ddg job_status: Passed torchscript_onnx_qnn: - inference_time: 7729.0 - throughput: 129.3828438349075 + inference_time: 5775.0 + throughput: 173.16017316017317 estimated_peak_memory_range: min: 28672 - max: 1399264 + max: 1702880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 377 - job_id: j0px008jp + job_id: j2p0xv9ep job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:13:10Z' + timestamp: '2024-08-27T11:52:35Z' - torchscript_onnx_tflite: - inference_time: 13170.0 - throughput: 75.9301442672741 + inference_time: 6742.0 + throughput: 148.3239394838327 estimated_peak_memory_range: - min: 53248 - max: 170561376 + min: 16384 + max: 154736272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 576 - job_id: j1glww9ep + job_id: j7gj8w385 job_status: Passed torchscript_onnx_qnn: - inference_time: 8444.0 - throughput: 118.42728564661299 + inference_time: 6632.0 + throughput: 150.78407720144753 estimated_peak_memory_range: - min: 12288 - max: 49613904 + min: 20480 + max: 60733248 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 377 - job_id: jogk661y5 + job_id: j1pvn9mmg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,28 +225,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:13:18Z' + timestamp: '2024-08-27T11:52:42Z' - torchscript_onnx_tflite: - inference_time: 11692.0 - throughput: 85.52856654122476 + inference_time: 5835.0 + throughput: 171.3796058269066 estimated_peak_memory_range: - min: 16384 - max: 8899472 + min: 20480 + max: 2634440 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 573 - layers_on_gpu: 1 + layers_on_npu: 574 + layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 576 - job_id: j1p3oolxp + job_id: jygz04z65 job_status: Passed torchscript_onnx_qnn: - inference_time: 7775.0 - throughput: 128.61736334405145 + inference_time: 5824.0 + throughput: 171.7032967032967 estimated_peak_memory_range: - min: 77824 - max: 1635952 + min: 24576 + max: 1280216 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 377 - job_id: jegn114vp + job_id: jogkk90og job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,28 +263,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:13:12Z' + timestamp: '2024-08-27T11:52:37Z' - torchscript_onnx_tflite: - inference_time: 11635.0 - throughput: 85.94757198109153 + inference_time: 5762.0 + throughput: 173.55085039916696 estimated_peak_memory_range: - min: 16384 - max: 8991728 + min: 24576 + max: 2279744 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 573 - layers_on_gpu: 1 + layers_on_npu: 574 + layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 576 - job_id: j1pv22y7g + job_id: jnp1mvolp job_status: Passed torchscript_onnx_qnn: - inference_time: 7703.0 - throughput: 129.81955082435414 + inference_time: 5756.0 + throughput: 173.73175816539262 estimated_peak_memory_range: - min: 102400 - max: 1394504 + min: 45056 + max: 1346192 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 377 - job_id: jep2oo1xg + job_id: j1glq18lp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,28 +301,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:13:14Z' + timestamp: '2024-08-27T11:52:39Z' - torchscript_onnx_tflite: - inference_time: 11819.0 - throughput: 84.60952703274388 + inference_time: 5758.0 + throughput: 173.67141368530739 estimated_peak_memory_range: - min: 16384 - max: 7977632 + min: 24576 + max: 2442024 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 573 - layers_on_gpu: 1 + layers_on_npu: 574 + layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 576 - job_id: jlpe6607g + job_id: jz5787nrp job_status: Passed torchscript_onnx_qnn: - inference_time: 7737.0 - throughput: 129.24906294429366 + inference_time: 5826.0 + throughput: 171.64435290078956 estimated_peak_memory_range: min: 61440 - max: 1544760 + max: 1599400 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 377 - job_id: j2p0oow2p + job_id: j1p3rw7zp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:13:16Z' + timestamp: '2024-08-27T11:52:40Z' - torchscript_onnx_qnn: - inference_time: 8291.0 - throughput: 120.61271257990592 + inference_time: 6164.0 + throughput: 162.23231667748215 estimated_peak_memory_range: - min: 163840 - max: 163840 + min: 159744 + max: 159744 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 377 - job_id: jz57oo1qg + job_id: jep2zvdmp job_status: Passed torchscript_onnx: - inference_time: 32465.0 - throughput: 30.80240258740182 + inference_time: 31950.0 + throughput: 31.29890453834116 estimated_peak_memory_range: - min: 133279744 - max: 133279744 + min: 132759552 + max: 132759552 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 389 - job_id: j1pv2287g + job_id: jnp1mv1lp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,30 +377,30 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:13:24Z' + timestamp: '2024-08-27T11:52:47Z' - name: CLIPImageEncoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 65346.0 - throughput: 15.303155510666299 + inference_time: 39036.0 + throughput: 25.617378829798135 estimated_peak_memory_range: - min: 86016 - max: 9946408 + min: 61440 + max: 2813992 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 574 - layers_on_gpu: 1 + layers_on_npu: 575 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 575 - job_id: jqpy88krg + job_id: j1glq1wlp job_status: Passed torchscript_onnx_qnn: - inference_time: 50085.0 - throughput: 19.96605770190676 + inference_time: 39047.0 + throughput: 25.610162112326172 estimated_peak_memory_range: - min: 131072 - max: 64295712 + min: 57344 + max: 56644000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -408,14 +408,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 369 - job_id: jmg9oo7qg + job_id: jo5mldkqg job_status: Passed torchscript_onnx: - inference_time: 174170.0 - throughput: 5.741516908767296 + inference_time: 165432.0 + throughput: 6.044779728226703 estimated_peak_memory_range: - min: 147456 - max: 204173664 + min: 151552 + max: 204535312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -423,7 +423,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 382 - job_id: jw56ooxv5 + job_id: jygz04y65 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -432,28 +432,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:13:21Z' + timestamp: '2024-08-27T11:52:44Z' - torchscript_onnx_tflite: - inference_time: 49170.0 - throughput: 20.33760423022168 + inference_time: 29092.0 + throughput: 34.37371098583803 estimated_peak_memory_range: - min: 81920 - max: 742057024 + min: 57344 + max: 488529824 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 574 - layers_on_gpu: 1 + layers_on_npu: 575 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 575 - job_id: j1p8jjdz5 + job_id: j1p3rwozp job_status: Passed torchscript_onnx_qnn: - inference_time: 37906.0 - throughput: 26.38104785522081 + inference_time: 29368.0 + throughput: 34.05066739308091 estimated_peak_memory_range: - min: 659456 - max: 99517184 + min: 667648 + max: 137281216 primary_compute_unit: NPU precision: fp16 layer_info: @@ -461,14 +461,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 369 - job_id: jvgd66ykp + job_id: jopr7ndeg job_status: Passed torchscript_onnx: - inference_time: 125426.0 - throughput: 7.972828600130755 + inference_time: 123736.0 + throughput: 8.081722376672916 estimated_peak_memory_range: - min: 798720 - max: 2980118208 + min: 876544 + max: 3057097648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -476,7 +476,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 382 - job_id: jwgoddx45 + job_id: jmg9qx2vp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -485,28 +485,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:13:23Z' + timestamp: '2024-08-27T11:52:46Z' - torchscript_onnx_tflite: - inference_time: 65399.0 - throughput: 15.29075368124895 + inference_time: 38354.0 + throughput: 26.07289982791886 estimated_peak_memory_range: - min: 94208 - max: 8936320 + min: 2805760 + max: 5444488 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 574 - layers_on_gpu: 1 + layers_on_npu: 575 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 575 - job_id: jn5q44x7g + job_id: j1pvn92mg job_status: Passed torchscript_onnx_qnn: - inference_time: 35624.0 - throughput: 28.070963395463732 + inference_time: 32485.0 + throughput: 30.783438510081577 estimated_peak_memory_range: - min: 733184 - max: 1913376 + min: 700416 + max: 2014160 primary_compute_unit: NPU precision: fp16 layer_info: @@ -514,7 +514,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 369 - job_id: jo5m991yg + job_id: j1p8k4r8p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -523,13 +523,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:13:10Z' + timestamp: '2024-08-27T11:52:36Z' - torchscript_onnx_tflite: - inference_time: 73910.0 - throughput: 13.529968881071573 + inference_time: 43693.0 + throughput: 22.886961298148446 estimated_peak_memory_range: - min: 569344 - max: 586647568 + min: 61440 + max: 472280896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -537,14 +537,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 575 - job_id: jw56oo9v5 + job_id: jlpenl60p job_status: Passed torchscript_onnx_qnn: - inference_time: 57122.0 - throughput: 17.506389832288786 + inference_time: 44949.0 + throughput: 22.24743598300296 estimated_peak_memory_range: min: 0 - max: 104086448 + max: 136401520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -552,7 +552,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 369 - job_id: jn5q44n7g + job_id: j7gj8wy85 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -561,28 +561,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:13:19Z' + timestamp: '2024-08-27T11:52:43Z' - torchscript_onnx_tflite: - inference_time: 66405.0 - throughput: 15.059106994955199 + inference_time: 38665.0 + throughput: 25.8631837579206 estimated_peak_memory_range: - min: 98304 - max: 8436552 + min: 0 + max: 1816140864 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 574 - layers_on_gpu: 1 + layers_on_npu: 575 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 575 - job_id: jwgodd745 + job_id: jz5wr1yjp job_status: Passed torchscript_onnx_qnn: - inference_time: 35557.0 - throughput: 28.12385746829035 + inference_time: 32530.0 + throughput: 30.740854595757764 estimated_peak_memory_range: - min: 696320 - max: 2031992 + min: 745472 + max: 1921584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -590,7 +590,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 369 - job_id: joprxxrvp + job_id: jn5qdm1mg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -599,28 +599,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:13:12Z' + timestamp: '2024-08-27T11:52:37Z' - torchscript_onnx_tflite: - inference_time: 65799.0 - throughput: 15.197799358652867 + inference_time: 38388.0 + throughput: 26.049807231426488 estimated_peak_memory_range: - min: 102400 - max: 7446576 + min: 90112 + max: 2886416 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 574 - layers_on_gpu: 1 + layers_on_npu: 575 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 575 - job_id: j7gj3367p + job_id: jvgdmz6lg job_status: Passed torchscript_onnx_qnn: - inference_time: 35653.0 - throughput: 28.048130592096037 + inference_time: 32680.0 + throughput: 30.599755201958384 estimated_peak_memory_range: - min: 737280 - max: 2127944 + min: 716800 + max: 2267464 primary_compute_unit: NPU precision: fp16 layer_info: @@ -628,7 +628,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 369 - job_id: jqpy88lrg + job_id: jw560dm75 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -637,28 +637,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:13:15Z' + timestamp: '2024-08-27T11:52:39Z' - torchscript_onnx_tflite: - inference_time: 66324.0 - throughput: 15.077498341475183 + inference_time: 38288.0 + throughput: 26.117843710823234 estimated_peak_memory_range: - min: 20480 - max: 1918917048 + min: 77824 + max: 2428336 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 574 - layers_on_gpu: 1 + layers_on_npu: 575 + layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 575 - job_id: jygzzzqzg + job_id: jqp4294lg job_status: Passed torchscript_onnx_qnn: - inference_time: 35517.0 - throughput: 28.155531154095222 + inference_time: 32624.0 + throughput: 30.65228052967141 estimated_peak_memory_range: - min: 712704 - max: 2111696 + min: 749568 + max: 1972280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -666,7 +666,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 369 - job_id: j1p8jjnz5 + job_id: jwgo94wdg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -675,10 +675,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:13:17Z' + timestamp: '2024-08-27T11:52:41Z' - torchscript_onnx_qnn: - inference_time: 36007.0 - throughput: 27.772377593245757 + inference_time: 31391.0 + throughput: 31.856264534420692 estimated_peak_memory_range: min: 602112 max: 602112 @@ -689,14 +689,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 369 - job_id: jqp4ee6qg + job_id: jqpyy724p job_status: Passed torchscript_onnx: - inference_time: 170594.0 - throughput: 5.861870874708372 + inference_time: 160701.0 + throughput: 6.222736635117392 estimated_peak_memory_range: - min: 198303744 - max: 198303744 + min: 196861952 + max: 196861952 primary_compute_unit: NPU precision: fp16 layer_info: @@ -704,7 +704,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 382 - job_id: j7gj3397p + job_id: jvgdmz4lg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -713,4 +713,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:13:25Z' + timestamp: '2024-08-27T11:52:48Z' diff --git a/qai_hub_models/models/openpose/perf.yaml b/qai_hub_models/models/openpose/perf.yaml index a12d43e9..f7432cc7 100644 --- a/qai_hub_models/models/openpose/perf.yaml +++ b/qai_hub_models/models/openpose/perf.yaml @@ -45,11 +45,11 @@ models: - name: OpenPose performance_metrics: - torchscript_onnx_tflite: - inference_time: 11679.0 - throughput: 85.62376915831835 + inference_time: 11701.0 + throughput: 85.46278095889241 estimated_peak_memory_range: - min: 196608 - max: 2409960 + min: 16384 + max: 1967600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: j2p0oox2p + job_id: jqpyyv60p job_status: Passed torchscript_onnx_qnn: - inference_time: 11785.0 - throughput: 84.85362749257531 + inference_time: 11858.0 + throughput: 84.33125316242199 estimated_peak_memory_range: - min: 565248 - max: 229794320 + min: 643072 + max: 240658496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jwgodd945 + job_id: j1p3rqmmp job_status: Passed torchscript_onnx: - inference_time: 11963.0 - throughput: 83.59107247345983 + inference_time: 12030.0 + throughput: 83.12551953449709 estimated_peak_memory_range: - min: 16384 - max: 119165528 + min: 32768 + max: 119051096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 189 - job_id: jvgd66mkp + job_id: jnp1mq8np job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:12:00Z' + timestamp: '2024-08-27T00:26:58Z' - torchscript_onnx_tflite: - inference_time: 8712.0 - throughput: 114.7842056932966 + inference_time: 8703.0 + throughput: 114.9029070435482 estimated_peak_memory_range: - min: 16384 - max: 38948256 + min: 208896 + max: 40353376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: j1p8jjkz5 + job_id: j2p0xer0p job_status: Passed torchscript_onnx_qnn: - inference_time: 8750.0 - throughput: 114.28571428571429 + inference_time: 8771.0 + throughput: 114.01208528103979 estimated_peak_memory_range: min: 618496 - max: 18769136 + max: 18233600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: j1pv22n7g + job_id: jwgo9ev1g job_status: Passed torchscript_onnx: - inference_time: 8914.0 - throughput: 112.1830827911151 + inference_time: 8902.0 + throughput: 112.33430689732644 estimated_peak_memory_range: - min: 208896 - max: 43535008 + min: 1126400 + max: 45559840 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 189 - job_id: jz57oo8qg + job_id: jvgdm7v6g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:12:01Z' + timestamp: '2024-08-27T00:26:59Z' - torchscript_onnx_tflite: - inference_time: 11692.0 - throughput: 85.52856654122476 + inference_time: 11679.0 + throughput: 85.62376915831835 estimated_peak_memory_range: - min: 200704 - max: 2688824 + min: 204800 + max: 2426192 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jogk66ky5 + job_id: j1p8kw7qp job_status: Passed torchscript_onnx_qnn: - inference_time: 11653.0 - throughput: 85.81481163648846 + inference_time: 11612.0 + throughput: 86.11780916293489 estimated_peak_memory_range: min: 634880 - max: 2577384 + max: 2053376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jlpe66n7g + job_id: j7gj8kl15 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:11:54Z' + timestamp: '2024-08-27T00:26:53Z' - torchscript_onnx_tflite: - inference_time: 23490.0 - throughput: 42.57130693912303 + inference_time: 23451.0 + throughput: 42.64210481429363 estimated_peak_memory_range: - min: 217088 - max: 40650784 + min: 212992 + max: 40336000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jn5q44d7g + job_id: jogkkryvg job_status: Passed torchscript_onnx_qnn: - inference_time: 23879.0 - throughput: 41.877800577913646 + inference_time: 23642.0 + throughput: 42.29760595550292 estimated_peak_memory_range: - min: 638976 - max: 19489360 + min: 618496 + max: 20393696 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jnp1oomk5 + job_id: jmg9q94mp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:11:59Z' + timestamp: '2024-08-27T00:26:57Z' - torchscript_onnx_tflite: - inference_time: 11752.0 - throughput: 85.0918992511913 + inference_time: 11659.0 + throughput: 85.77064928381508 estimated_peak_memory_range: - min: 192512 - max: 2504456 + min: 20480 + max: 2423376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: j1glwwqep + job_id: jn5qd92eg job_status: Passed torchscript_onnx_qnn: - inference_time: 11748.0 - throughput: 85.12087163772557 + inference_time: 11612.0 + throughput: 86.11780916293489 estimated_peak_memory_range: - min: 679936 - max: 2293224 + min: 688128 + max: 2452648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jygzzz0zg + job_id: jlpen4v8p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:11:55Z' + timestamp: '2024-08-27T00:26:54Z' - torchscript_onnx_tflite: inference_time: 11703.0 throughput: 85.4481756814492 estimated_peak_memory_range: min: 200704 - max: 1730392 + max: 2186848 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: jw56oo0v5 + job_id: j1glqek2p job_status: Passed torchscript_onnx_qnn: - inference_time: 11669.0 - throughput: 85.69714628502871 + inference_time: 11615.0 + throughput: 86.09556607834696 estimated_peak_memory_range: - min: 655360 - max: 2232744 + min: 671744 + max: 2293320 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jz5wyyrzg + job_id: jygz0v745 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:11:56Z' + timestamp: '2024-08-27T00:26:55Z' - torchscript_onnx_tflite: - inference_time: 11713.0 - throughput: 85.37522410996328 + inference_time: 11761.0 + throughput: 85.02678343678258 estimated_peak_memory_range: - min: 208896 - max: 1941272 + min: 200704 + max: 2344192 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 103 - job_id: j1p3oorxp + job_id: jw560q1n5 job_status: Passed torchscript_onnx_qnn: - inference_time: 11708.0 - throughput: 85.41168431841476 + inference_time: 11713.0 + throughput: 85.37522410996328 estimated_peak_memory_range: - min: 688128 - max: 1955880 + min: 684032 + max: 1834712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jmg9ooqqg + job_id: jz5wrm94p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:11:58Z' + timestamp: '2024-08-27T00:26:56Z' - torchscript_onnx_qnn: - inference_time: 12461.0 - throughput: 80.25038118931064 + inference_time: 12373.0 + throughput: 80.82114281095934 estimated_peak_memory_range: - min: 602112 - max: 602112 + min: 655360 + max: 655360 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: j7gj3387p + job_id: j1pvnzwzg job_status: Passed torchscript_onnx: - inference_time: 12379.0 - throughput: 80.78196946441554 + inference_time: 12384.0 + throughput: 80.74935400516796 estimated_peak_memory_range: - min: 107540480 - max: 107540480 + min: 107737088 + max: 107737088 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 189 - job_id: jqp4ee2qg + job_id: jz578vdnp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:12:02Z' + timestamp: '2024-08-27T00:27:00Z' diff --git a/qai_hub_models/models/posenet_mobilenet/perf.yaml b/qai_hub_models/models/posenet_mobilenet/perf.yaml index 4e972f9d..a30dd530 100644 --- a/qai_hub_models/models/posenet_mobilenet/perf.yaml +++ b/qai_hub_models/models/posenet_mobilenet/perf.yaml @@ -45,11 +45,11 @@ models: - name: Posenet-Mobilenet performance_metrics: - torchscript_onnx_tflite: - inference_time: 1399.0 - throughput: 714.7962830593281 + inference_time: 1365.0 + throughput: 732.6007326007326 estimated_peak_memory_range: - min: 12288 - max: 1434048 + min: 24576 + max: 2705152 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j1p3oo7mp + job_id: jqpyyvj0p job_status: Passed torchscript_onnx_qnn: - inference_time: 1449.0 - throughput: 690.1311249137336 + inference_time: 1446.0 + throughput: 691.5629322268327 estimated_peak_memory_range: - min: 12288 - max: 23445360 + min: 1605632 + max: 12385280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 69 - job_id: jmg9oo2mg + job_id: j1p3rqemp job_status: Passed torchscript_onnx: - inference_time: 2081.0 - throughput: 480.5382027871216 + inference_time: 2117.0 + throughput: 472.3665564478035 estimated_peak_memory_range: - min: 12288 - max: 8002728 + min: 16384 + max: 8302592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jqp4ee4qg + job_id: jnp1mqenp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:09:34Z' + timestamp: '2024-08-27T00:24:38Z' - torchscript_onnx_tflite: - inference_time: 975.0 - throughput: 1025.6410256410256 + inference_time: 959.0 + throughput: 1042.752867570386 estimated_peak_memory_range: - min: 12288 - max: 39925712 + min: 16384 + max: 39650176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jwgoddw15 + job_id: j2p0xel0p job_status: Passed torchscript_onnx_qnn: - inference_time: 1019.0 - throughput: 981.3542688910696 + inference_time: 1023.0 + throughput: 977.5171065493646 estimated_peak_memory_range: min: 0 - max: 13051056 + max: 14554656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 69 - job_id: jnp1oo1n5 + job_id: jwgo9e31g job_status: Passed torchscript_onnx: - inference_time: 1439.0 - throughput: 694.9270326615705 + inference_time: 1431.0 + throughput: 698.8120195667366 estimated_peak_memory_range: - min: 475136 - max: 43775824 + min: 0 + max: 43765808 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: j0px00rjp + job_id: jvgdm7o6g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:09:35Z' + timestamp: '2024-08-27T00:24:39Z' - torchscript_onnx_tflite: - inference_time: 1397.0 - throughput: 715.8196134574088 + inference_time: 1401.0 + throughput: 713.7758743754462 estimated_peak_memory_range: min: 12288 - max: 1372952 + max: 1363184 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j1pv22mzg + job_id: j1p8kwzqp job_status: Passed torchscript_onnx_qnn: - inference_time: 1374.0 - throughput: 727.802037845706 + inference_time: 1381.0 + throughput: 724.112961622013 estimated_peak_memory_range: - min: 1617920 - max: 3396912 + min: 1622016 + max: 2761584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 69 - job_id: jz5wyyzzg + job_id: j7gj8ke15 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:09:28Z' + timestamp: '2024-08-27T00:24:33Z' - torchscript_onnx_tflite: - inference_time: 2249.0 - throughput: 444.642063139173 + inference_time: 2185.0 + throughput: 457.66590389016017 estimated_peak_memory_range: - min: 16384 - max: 41038352 + min: 315392 + max: 42377440 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j7gj33y1p + job_id: jogkkr3vg job_status: Passed torchscript_onnx_qnn: - inference_time: 2282.0 - throughput: 438.21209465381247 + inference_time: 2257.0 + throughput: 443.06601683650865 estimated_peak_memory_range: min: 1597440 - max: 21542128 + max: 19171040 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 69 - job_id: jz57oonqg + job_id: jmg9q9wmp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:09:33Z' + timestamp: '2024-08-27T00:24:37Z' - torchscript_onnx_tflite: - inference_time: 1398.0 - throughput: 715.307582260372 + inference_time: 1364.0 + throughput: 733.1378299120234 estimated_peak_memory_range: - min: 24576 - max: 1348552 + min: 12288 + max: 1577656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jlpe66x8g + job_id: jn5qd93eg job_status: Passed torchscript_onnx_qnn: - inference_time: 1383.0 - throughput: 723.0657989877079 + inference_time: 1392.0 + throughput: 718.3908045977012 estimated_peak_memory_range: - min: 1626112 - max: 3201776 + min: 1634304 + max: 3498240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 69 - job_id: jmg9oo2qg + job_id: jlpen4k8p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:09:29Z' + timestamp: '2024-08-27T00:24:34Z' - torchscript_onnx_tflite: - inference_time: 1403.0 - throughput: 712.7583749109052 + inference_time: 1363.0 + throughput: 733.6757153338225 estimated_peak_memory_range: - min: 28672 - max: 42092224 + min: 12288 + max: 13156592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jygzzzy4g + job_id: j1glqe32p job_status: Passed torchscript_onnx_qnn: - inference_time: 1385.0 - throughput: 722.0216606498195 + inference_time: 1386.0 + throughput: 721.5007215007215 estimated_peak_memory_range: - min: 1626112 - max: 3336608 + min: 1613824 + max: 2997232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 69 - job_id: jnp1oo1k5 + job_id: jygz0vr45 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:09:30Z' + timestamp: '2024-08-27T00:24:35Z' - torchscript_onnx_tflite: - inference_time: 1398.0 - throughput: 715.307582260372 + inference_time: 1365.0 + throughput: 732.6007326007326 estimated_peak_memory_range: min: 12288 - max: 9739144 + max: 9630752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jz5wyyz4g + job_id: jw560qnn5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1401.0 - throughput: 713.7758743754462 + inference_time: 1387.0 + throughput: 720.9805335255949 estimated_peak_memory_range: min: 1613824 - max: 2923440 + max: 3211608 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 69 - job_id: jvgd664kp + job_id: jz5wrmq4p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:09:31Z' + timestamp: '2024-08-27T00:24:36Z' - torchscript_onnx_qnn: - inference_time: 1562.0 - throughput: 640.2048655569782 + inference_time: 1773.0 + throughput: 564.0157924421884 estimated_peak_memory_range: min: 1589248 max: 1589248 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 69 - job_id: jvgd6646p + job_id: j1pvnzvzg job_status: Passed torchscript_onnx: - inference_time: 2183.0 - throughput: 458.0852038479157 + inference_time: 2208.0 + throughput: 452.8985507246377 estimated_peak_memory_range: - min: 7929856 - max: 7929856 + min: 9535488 + max: 9535488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jo5m99kyg + job_id: jz578vxnp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:09:36Z' + timestamp: '2024-08-27T00:24:40Z' diff --git a/qai_hub_models/models/posenet_mobilenet_quantized/perf.yaml b/qai_hub_models/models/posenet_mobilenet_quantized/perf.yaml index 1ea7dbde..ff37bb23 100644 --- a/qai_hub_models/models/posenet_mobilenet_quantized/perf.yaml +++ b/qai_hub_models/models/posenet_mobilenet_quantized/perf.yaml @@ -48,11 +48,11 @@ models: - name: Posenet-Mobilenet-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 560.0 - throughput: 1785.7142857142858 + inference_time: 554.0 + throughput: 1805.0541516245487 estimated_peak_memory_range: min: 12288 - max: 105229992 + max: 6432616 primary_compute_unit: NPU precision: int8 layer_info: @@ -60,14 +60,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 48 - job_id: j1p3ooomp + job_id: jep2z306p job_status: Passed torchscript_onnx_qnn: - inference_time: 624.0 - throughput: 1602.5641025641025 + inference_time: 629.0 + throughput: 1589.825119236884 estimated_peak_memory_range: - min: 16384 - max: 7655712 + min: 12288 + max: 108492928 primary_compute_unit: NPU precision: int8 layer_info: @@ -75,7 +75,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jvgd6666p + job_id: j1pvnzrzg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -84,13 +84,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:08:46Z' + timestamp: '2024-08-27T00:23:55Z' - torchscript_onnx_tflite: - inference_time: 395.0 - throughput: 2531.6455696202534 + inference_time: 397.0 + throughput: 2518.891687657431 estimated_peak_memory_range: min: 12288 - max: 48243200 + max: 49076816 primary_compute_unit: NPU precision: int8 layer_info: @@ -98,14 +98,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 48 - job_id: jwgoddd15 + job_id: jqpyyvr0p job_status: Passed torchscript_onnx_qnn: - inference_time: 444.0 - throughput: 2252.252252252252 + inference_time: 441.0 + throughput: 2267.573696145125 estimated_peak_memory_range: min: 0 - max: 16907408 + max: 18101968 primary_compute_unit: NPU precision: int8 layer_info: @@ -113,7 +113,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jz57ooong + job_id: j7gj8k215 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -122,13 +122,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:08:47Z' + timestamp: '2024-08-27T00:23:56Z' - torchscript_onnx_tflite: - inference_time: 564.0 - throughput: 1773.049645390071 + inference_time: 553.0 + throughput: 1808.3182640144666 estimated_peak_memory_range: min: 12288 - max: 1290232 + max: 1268824 primary_compute_unit: NPU precision: int8 layer_info: @@ -136,14 +136,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 48 - job_id: j1pv222zg + job_id: j2p0xe20p job_status: Passed torchscript_onnx_qnn: - inference_time: 555.0 - throughput: 1801.8018018018017 + inference_time: 558.0 + throughput: 1792.1146953405018 estimated_peak_memory_range: min: 425984 - max: 1651192 + max: 1618192 primary_compute_unit: NPU precision: int8 layer_info: @@ -151,7 +151,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: j0px0008p + job_id: jygz0vj45 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -160,13 +160,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:08:50Z' + timestamp: '2024-08-27T00:23:58Z' - torchscript_onnx_tflite: - inference_time: 759.0 - throughput: 1317.5230566534915 + inference_time: 720.0 + throughput: 1388.888888888889 estimated_peak_memory_range: min: 12288 - max: 49095872 + max: 50424144 primary_compute_unit: NPU precision: int8 layer_info: @@ -174,14 +174,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 48 - job_id: j7gj3331p + job_id: j1p8kwmqp job_status: Passed torchscript_onnx_qnn: - inference_time: 826.0 - throughput: 1210.6537530266344 + inference_time: 792.0 + throughput: 1262.6262626262626 estimated_peak_memory_range: min: 413696 - max: 21362400 + max: 21281200 primary_compute_unit: NPU precision: int8 layer_info: @@ -189,7 +189,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jep2ooo6g + job_id: jvgdm7q6g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -198,13 +198,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:08:54Z' + timestamp: '2024-08-27T00:24:03Z' - torchscript_onnx_tflite: - inference_time: 584.0 - throughput: 1712.3287671232877 + inference_time: 553.0 + throughput: 1808.3182640144666 estimated_peak_memory_range: - min: 16384 - max: 1380488 + min: 12288 + max: 64377104 primary_compute_unit: NPU precision: int8 layer_info: @@ -212,14 +212,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 48 - job_id: jlpe6668g + job_id: jogkkrqvg job_status: Passed torchscript_onnx_qnn: inference_time: 561.0 throughput: 1782.5311942959001 estimated_peak_memory_range: min: 425984 - max: 1654024 + max: 2217632 primary_compute_unit: NPU precision: int8 layer_info: @@ -227,7 +227,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jo5m9997g + job_id: jz5wrm34p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -236,13 +236,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:08:51Z' + timestamp: '2024-08-27T00:23:59Z' - torchscript_onnx_tflite: - inference_time: 582.0 - throughput: 1718.213058419244 + inference_time: 548.0 + throughput: 1824.8175182481752 estimated_peak_memory_range: - min: 16384 - max: 1442984 + min: 12288 + max: 66237624 primary_compute_unit: NPU precision: int8 layer_info: @@ -250,14 +250,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 48 - job_id: jygzzzz4g + job_id: jn5qd9reg job_status: Passed torchscript_onnx_qnn: - inference_time: 562.0 - throughput: 1779.3594306049822 + inference_time: 564.0 + throughput: 1773.049645390071 estimated_peak_memory_range: - min: 430080 - max: 1804688 + min: 421888 + max: 2157400 primary_compute_unit: NPU precision: int8 layer_info: @@ -265,7 +265,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jegn111jp + job_id: jmg9q9ymp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -274,13 +274,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:08:52Z' + timestamp: '2024-08-27T00:24:00Z' - torchscript_onnx_tflite: - inference_time: 584.0 - throughput: 1712.3287671232877 + inference_time: 556.0 + throughput: 1798.5611510791366 estimated_peak_memory_range: min: 12288 - max: 1345144 + max: 9048872 primary_compute_unit: NPU precision: int8 layer_info: @@ -288,14 +288,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 48 - job_id: jz5wyyy4g + job_id: j1glqe22p job_status: Passed torchscript_onnx_qnn: - inference_time: 566.0 - throughput: 1766.7844522968198 + inference_time: 556.0 + throughput: 1798.5611510791366 estimated_peak_memory_range: min: 430080 - max: 1746104 + max: 1672440 primary_compute_unit: NPU precision: int8 layer_info: @@ -303,7 +303,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: joprxxxkp + job_id: jnp1mqwnp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -312,13 +312,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:08:53Z' + timestamp: '2024-08-27T00:24:02Z' - torchscript_onnx_tflite: - inference_time: 2212.0 - throughput: 452.07956600361666 + inference_time: 2131.0 + throughput: 469.2632566870014 estimated_peak_memory_range: - min: 36864 - max: 27250448 + min: 12288 + max: 28329984 primary_compute_unit: NPU precision: int8 layer_info: @@ -326,14 +326,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 48 - job_id: jmg9ooomg + job_id: j1p3rq1mp job_status: Passed torchscript_onnx_qnn: - inference_time: 2834.0 - throughput: 352.85815102328866 + inference_time: 2852.0 + throughput: 350.6311360448808 estimated_peak_memory_range: - min: 16384 - max: 8663200 + min: 12288 + max: 8625136 primary_compute_unit: NPU precision: int8 layer_info: @@ -341,7 +341,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jqpy8880g + job_id: jz578vlnp job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -350,13 +350,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:08:55Z' + timestamp: '2024-08-27T00:24:03Z' - torchscript_onnx_tflite: - inference_time: 13315.0 - throughput: 75.10326699211416 + inference_time: 13054.0 + throughput: 76.60487206986365 estimated_peak_memory_range: - min: 450560 - max: 7112584 + min: 90112 + max: 18716936 primary_compute_unit: NPU precision: int8 layer_info: @@ -364,7 +364,7 @@ models: layers_on_gpu: 3 layers_on_cpu: 0 total_layers: 48 - job_id: jnp1ooon5 + job_id: jwgo9en1g job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -373,13 +373,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:08:44Z' + timestamp: '2024-08-27T00:23:54Z' - torchscript_onnx_qnn: - inference_time: 732.0 - throughput: 1366.120218579235 + inference_time: 694.0 + throughput: 1440.922190201729 estimated_peak_memory_range: - min: 1413120 - max: 1413120 + min: 397312 + max: 397312 primary_compute_unit: NPU precision: int8 layer_info: @@ -387,7 +387,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jqp4eee2g + job_id: jlpen4w8p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -396,4 +396,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:08:48Z' + timestamp: '2024-08-27T00:23:57Z' diff --git a/qai_hub_models/models/quicksrnetlarge/perf.yaml b/qai_hub_models/models/quicksrnetlarge/perf.yaml index ea73e2c9..04351c48 100644 --- a/qai_hub_models/models/quicksrnetlarge/perf.yaml +++ b/qai_hub_models/models/quicksrnetlarge/perf.yaml @@ -45,11 +45,11 @@ models: - name: QuickSRNetLarge performance_metrics: - torchscript_onnx_tflite: - inference_time: 2468.0 - throughput: 405.1863857374392 + inference_time: 2474.0 + throughput: 404.2037186742118 estimated_peak_memory_range: min: 28672 - max: 20205016 + max: 1511992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 31 - job_id: j1pv22lzg + job_id: j7gj8wr75 job_status: Passed torchscript_onnx_qnn: - inference_time: 2123.0 - throughput: 471.03155911446066 + inference_time: 2116.0 + throughput: 472.5897920604915 estimated_peak_memory_range: - min: 90112 - max: 3387472 + min: 86016 + max: 3307544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jvgd66d6p + job_id: jz5wr1ljp job_status: Passed torchscript_onnx: - inference_time: 2659.0 - throughput: 376.081233546446 + inference_time: 2649.0 + throughput: 377.5009437523594 estimated_peak_memory_range: - min: 12288 - max: 15580248 + min: 16384 + max: 2621104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 33 - job_id: jqpy88o0g + job_id: jegnw71mg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:08:10Z' + timestamp: '2024-08-27T11:51:33Z' - torchscript_onnx_tflite: - inference_time: 1733.0 - throughput: 577.0340450086555 + inference_time: 1681.0 + throughput: 594.883997620464 estimated_peak_memory_range: min: 0 - max: 31375520 + max: 31370432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 31 - job_id: j7gj33r1p + job_id: jlpenl77p job_status: Passed torchscript_onnx_qnn: - inference_time: 1514.0 - throughput: 660.5019815059445 + inference_time: 1503.0 + throughput: 665.335994677312 estimated_peak_memory_range: - min: 204800 - max: 12080864 + min: 208896 + max: 12818368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jz57ooeng + job_id: jmg9qxzvp job_status: Passed torchscript_onnx: - inference_time: 1804.0 - throughput: 554.3237250554324 + inference_time: 1813.0 + throughput: 551.5719801434087 estimated_peak_memory_range: min: 0 - max: 35019632 + max: 34479200 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 33 - job_id: j2p0ooo0p + job_id: jopr7nxeg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:08:11Z' + timestamp: '2024-08-27T11:51:34Z' - torchscript_onnx_tflite: - inference_time: 2452.0 - throughput: 407.8303425774878 + inference_time: 2430.0 + throughput: 411.52263374485597 estimated_peak_memory_range: - min: 28672 - max: 18448104 + min: 24576 + max: 1319144 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 31 - job_id: jlpe6678g + job_id: jygz04lz5 job_status: Passed torchscript_onnx_qnn: - inference_time: 2121.0 - throughput: 471.4757190004715 + inference_time: 2113.0 + throughput: 473.260766682442 estimated_peak_memory_range: - min: 229376 - max: 1811280 + min: 225280 + max: 1499432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: j0px00l8p + job_id: jvgdmzdlg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:08:04Z' + timestamp: '2024-08-27T11:51:29Z' - torchscript_onnx_tflite: - inference_time: 5456.0 - throughput: 183.28445747800586 + inference_time: 4012.0 + throughput: 249.25224327018944 estimated_peak_memory_range: - min: 12627968 - max: 45049088 + min: 12607488 + max: 45628992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 31 - job_id: jygzzzl4g + job_id: jz5wr1lzp job_status: Passed torchscript_onnx_qnn: - inference_time: 3436.0 - throughput: 291.0360884749709 + inference_time: 3494.0 + throughput: 286.20492272467084 estimated_peak_memory_range: min: 208896 - max: 15439968 + max: 14113184 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jep2oor6g + job_id: jo5mld9qg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:08:09Z' + timestamp: '2024-08-27T11:51:32Z' - torchscript_onnx_tflite: - inference_time: 2493.0 - throughput: 401.1231448054553 + inference_time: 2402.0 + throughput: 416.31973355537053 estimated_peak_memory_range: min: 24576 - max: 8888048 + max: 41625552 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 31 - job_id: jz5wyyl4g + job_id: jmg9qxzqp job_status: Passed torchscript_onnx_qnn: - inference_time: 2112.0 - throughput: 473.4848484848485 + inference_time: 2118.0 + throughput: 472.14353163361665 estimated_peak_memory_range: - min: 221184 - max: 1413960 + min: 233472 + max: 1512664 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jo5m9907g + job_id: jz5787orp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:08:06Z' + timestamp: '2024-08-27T11:51:30Z' - torchscript_onnx_tflite: - inference_time: 2433.0 - throughput: 411.0152075626798 + inference_time: 2464.0 + throughput: 405.84415584415586 estimated_peak_memory_range: - min: 1810432 - max: 81503616 + min: 24576 + max: 37134096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 31 - job_id: jmg9oozmg + job_id: jnp1mvnkp job_status: Passed torchscript_onnx_qnn: - inference_time: 2103.0 - throughput: 475.51117451260103 + inference_time: 2117.0 + throughput: 472.3665564478035 estimated_peak_memory_range: - min: 221184 - max: 1450384 + min: 229376 + max: 1950448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jegn11zjp + job_id: jqp429elg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:08:07Z' + timestamp: '2024-08-27T11:51:31Z' - torchscript_onnx_tflite: - inference_time: 2451.0 - throughput: 407.9967360261118 + inference_time: 2461.0 + throughput: 406.33888663145063 estimated_peak_memory_range: - min: 32768 - max: 1430344 + min: 20480 + max: 9176752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 31 - job_id: jnp1oonn5 + job_id: jvgdmzdkg job_status: Passed torchscript_onnx_qnn: - inference_time: 2123.0 - throughput: 471.03155911446066 + inference_time: 2121.0 + throughput: 471.4757190004715 estimated_peak_memory_range: - min: 229376 - max: 4794880 + min: 221184 + max: 1827896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: joprxxlkp + job_id: j0pxzd095 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:08:08Z' + timestamp: '2024-08-27T11:51:32Z' - torchscript_onnx_qnn: - inference_time: 2436.0 - throughput: 410.5090311986864 + inference_time: 2406.0 + throughput: 415.6275976724855 estimated_peak_memory_range: min: 217088 max: 217088 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jqp4eey2g + job_id: jnp1mvnlp job_status: Passed torchscript_onnx: - inference_time: 2726.0 - throughput: 366.83785766691125 + inference_time: 2706.0 + throughput: 369.5491500369549 estimated_peak_memory_range: - min: 8990720 - max: 8990720 + min: 9019392 + max: 9019392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 33 - job_id: j1p8jjjq5 + job_id: jep2zvomp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:08:12Z' + timestamp: '2024-08-27T11:51:35Z' diff --git a/qai_hub_models/models/quicksrnetlarge_quantized/perf.yaml b/qai_hub_models/models/quicksrnetlarge_quantized/perf.yaml index a1a8d15b..c3a377f0 100644 --- a/qai_hub_models/models/quicksrnetlarge_quantized/perf.yaml +++ b/qai_hub_models/models/quicksrnetlarge_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: QuickSRNetLarge-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 1494.0 - throughput: 669.3440428380187 + inference_time: 1428.0 + throughput: 700.2801120448179 estimated_peak_memory_range: - min: 3186688 - max: 16773408 + min: 12288 + max: 1640072 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: jwgoddr15 + job_id: jogkkreng job_status: Passed torchscript_onnx_qnn: - inference_time: 906.0 - throughput: 1103.7527593818984 + inference_time: 908.0 + throughput: 1101.3215859030836 estimated_peak_memory_range: min: 28672 - max: 8141328 + max: 8507688 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: jz57oo4ng + job_id: jygz0v2x5 job_status: Passed torchscript_onnx: - inference_time: 1066.0 - throughput: 938.0863039399625 + inference_time: 1045.0 + throughput: 956.9377990430622 estimated_peak_memory_range: min: 12288 - max: 65064776 + max: 16937136 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 22 - job_id: j1p8jjeq5 + job_id: jegnwr0qg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:07:32Z' + timestamp: '2024-08-27T00:22:49Z' - torchscript_onnx_tflite: - inference_time: 1171.0 - throughput: 853.9709649871904 + inference_time: 1095.0 + throughput: 913.2420091324201 estimated_peak_memory_range: - min: 0 - max: 26834208 + min: 12288 + max: 28205568 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: j1pv22dzg + job_id: jn5qd96og job_status: Passed torchscript_onnx_qnn: - inference_time: 639.0 - throughput: 1564.9452269170579 + inference_time: 645.0 + throughput: 1550.3875968992247 estimated_peak_memory_range: min: 12288 - max: 14739200 + max: 13749728 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: jqp4ee12g + job_id: jz5wrmwmp job_status: Passed torchscript_onnx: - inference_time: 762.0 - throughput: 1312.3359580052493 + inference_time: 755.0 + throughput: 1324.5033112582782 estimated_peak_memory_range: min: 0 - max: 28632464 + max: 29894736 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 22 - job_id: jogk662v5 + job_id: jopr7167g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:07:33Z' + timestamp: '2024-08-27T00:22:49Z' - torchscript_onnx_tflite: - inference_time: 1922.0 - throughput: 520.2913631633714 + inference_time: 1913.0 + throughput: 522.7391531625718 estimated_peak_memory_range: - min: 806912 - max: 2226208 + min: 20480 + max: 1433344 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: j7gj3371p + job_id: j1glqe4mp job_status: Passed torchscript_onnx_qnn: - inference_time: 804.0 - throughput: 1243.7810945273632 + inference_time: 829.0 + throughput: 1206.2726176115802 estimated_peak_memory_range: - min: 81920 - max: 1252888 + min: 86016 + max: 1307712 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: jo5m99m7g + job_id: jnp1mq27p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:07:25Z' + timestamp: '2024-08-27T00:22:43Z' - torchscript_onnx_tflite: - inference_time: 2756.0 - throughput: 362.84470246734395 + inference_time: 1722.0 + throughput: 580.7200929152149 estimated_peak_memory_range: - min: 12288 - max: 29019344 + min: 1966080 + max: 32534784 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: jlpe66z8g + job_id: jw560q2y5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1051.0 - throughput: 951.4747859181732 + inference_time: 1069.0 + throughput: 935.4536950420954 estimated_peak_memory_range: min: 12288 - max: 14971776 + max: 14095232 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: jqpy88x0g + job_id: j0pxze9l5 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:07:30Z' + timestamp: '2024-08-27T00:22:47Z' - torchscript_onnx_tflite: - inference_time: 1484.0 - throughput: 673.8544474393531 + inference_time: 1450.0 + throughput: 689.6551724137931 estimated_peak_memory_range: - min: 24576 - max: 1358280 + min: 180224 + max: 14091936 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: jygzzzm4g + job_id: j1p3rqnnp job_status: Passed torchscript_onnx_qnn: - inference_time: 820.0 - throughput: 1219.5121951219512 + inference_time: 807.0 + throughput: 1239.1573729863692 estimated_peak_memory_range: - min: 81920 - max: 1361016 + min: 86016 + max: 1368896 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: jegn11njp + job_id: jvgdm7nzg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:07:26Z' + timestamp: '2024-08-27T00:22:44Z' - torchscript_onnx_tflite: - inference_time: 1503.0 - throughput: 665.335994677312 + inference_time: 1431.0 + throughput: 698.8120195667366 estimated_peak_memory_range: - min: 28672 - max: 80714848 + min: 53248 + max: 1443080 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: jz5wyy74g + job_id: jwgo9ezkg job_status: Passed torchscript_onnx_qnn: - inference_time: 809.0 - throughput: 1236.0939431396787 + inference_time: 808.0 + throughput: 1237.6237623762377 estimated_peak_memory_range: - min: 81920 - max: 1305952 + min: 90112 + max: 2214344 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: joprxx0kp + job_id: jz578v29p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:07:28Z' + timestamp: '2024-08-27T00:22:45Z' - torchscript_onnx_tflite: - inference_time: 1504.0 - throughput: 664.8936170212766 + inference_time: 1438.0 + throughput: 695.4102920723227 estimated_peak_memory_range: - min: 61440 - max: 1633568 + min: 24576 + max: 2494592 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: jmg9oommg + job_id: j1pvnzqrg job_status: Passed torchscript_onnx_qnn: - inference_time: 807.0 - throughput: 1239.1573729863692 + inference_time: 804.0 + throughput: 1243.7810945273632 estimated_peak_memory_range: min: 73728 - max: 1404136 + max: 1257304 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: jep2oow6g + job_id: jqp42jn1g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:07:29Z' + timestamp: '2024-08-27T00:22:46Z' - torchscript_onnx_tflite: - inference_time: 4173.0 - throughput: 239.63575365444524 + inference_time: 5822.0 + throughput: 171.76228100309172 estimated_peak_memory_range: - min: 3186688 - max: 22989680 + min: 3178496 + max: 24501536 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: jnp1oojn5 + job_id: j7gj8kde5 job_status: Passed torchscript_onnx_qnn: - inference_time: 3065.0 - throughput: 326.2642740619902 + inference_time: 3051.0 + throughput: 327.76138970829237 estimated_peak_memory_range: min: 65536 - max: 8216224 + max: 8187856 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: j2p0oom0p + job_id: jo5mlve9g job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:07:31Z' + timestamp: '2024-08-27T00:22:48Z' - torchscript_onnx_tflite: - inference_time: 38278.0 - throughput: 26.12466691049689 + inference_time: 38638.0 + throughput: 25.88125679382991 estimated_peak_memory_range: - min: 3321856 - max: 10090944 + min: 3448832 + max: 11036080 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 33 - job_id: jvgd6636p + job_id: jlpen4ovp job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:07:20Z' + timestamp: '2024-08-27T00:22:39Z' - torchscript_onnx_qnn: - inference_time: 1050.0 - throughput: 952.3809523809524 + inference_time: 991.0 + throughput: 1009.0817356205853 estimated_peak_memory_range: - min: 1126400 - max: 1126400 + min: 1048576 + max: 1048576 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 18 - job_id: j0px0048p + job_id: jmg9q908p job_status: Passed torchscript_onnx: - inference_time: 1182.0 - throughput: 846.0236886632825 + inference_time: 1067.0 + throughput: 937.207122774133 estimated_peak_memory_range: - min: 3436544 - max: 3436544 + min: 3395584 + max: 3395584 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 22 - job_id: jn5q44leg + job_id: jqpyyvzlp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:07:34Z' + timestamp: '2024-08-27T00:22:51Z' diff --git a/qai_hub_models/models/quicksrnetmedium/perf.yaml b/qai_hub_models/models/quicksrnetmedium/perf.yaml index ff6ffdeb..e0ff88db 100644 --- a/qai_hub_models/models/quicksrnetmedium/perf.yaml +++ b/qai_hub_models/models/quicksrnetmedium/perf.yaml @@ -45,11 +45,11 @@ models: - name: QuickSRNetMedium performance_metrics: - torchscript_onnx_tflite: - inference_time: 1446.0 - throughput: 691.5629322268327 + inference_time: 1360.0 + throughput: 735.2941176470588 estimated_peak_memory_range: - min: 12623872 - max: 19916192 + min: 12660736 + max: 79973296 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 17 - job_id: jmg9ood8g + job_id: j1glqevmp job_status: Passed torchscript_onnx_qnn: - inference_time: 1016.0 - throughput: 984.2519685039371 + inference_time: 994.0 + throughput: 1006.0362173038229 estimated_peak_memory_range: - min: 2125824 - max: 4419880 + min: 221184 + max: 73075720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 17 - job_id: jz57oo9ng + job_id: jygz0v3x5 job_status: Passed torchscript_onnx: - inference_time: 1568.0 - throughput: 637.7551020408164 + inference_time: 1563.0 + throughput: 639.7952655150352 estimated_peak_memory_range: - min: 212992 - max: 2994896 + min: 258048 + max: 1849232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 19 - job_id: j2p0ooj0p + job_id: jo5mlvn9g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:06:50Z' + timestamp: '2024-08-27T00:22:11Z' - torchscript_onnx_tflite: - inference_time: 924.0 - throughput: 1082.2510822510822 + inference_time: 873.0 + throughput: 1145.475372279496 estimated_peak_memory_range: min: 0 - max: 21516496 + max: 21991104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 17 - job_id: jnp1oo675 + job_id: jw560qyy5 job_status: Passed torchscript_onnx_qnn: - inference_time: 649.0 - throughput: 1540.8320493066255 + inference_time: 661.0 + throughput: 1512.8593040847202 estimated_peak_memory_range: - min: 204800 - max: 12195984 + min: 208896 + max: 11068880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 17 - job_id: jqp4ee32g + job_id: jz5wrmemp job_status: Passed torchscript_onnx: - inference_time: 1032.0 - throughput: 968.9922480620155 + inference_time: 1016.0 + throughput: 984.2519685039371 estimated_peak_memory_range: min: 0 - max: 23521584 + max: 23783632 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 19 - job_id: j1p8jjxq5 + job_id: jegnwr6qg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:06:51Z' + timestamp: '2024-08-27T00:22:12Z' - torchscript_onnx_tflite: - inference_time: 1428.0 - throughput: 700.2801120448179 + inference_time: 1342.0 + throughput: 745.156482861401 estimated_peak_memory_range: min: 28672 - max: 67013488 + max: 1279448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 17 - job_id: jvgd662zp + job_id: j1p3rqjnp job_status: Passed torchscript_onnx_qnn: - inference_time: 1017.0 - throughput: 983.284169124877 + inference_time: 1016.0 + throughput: 984.2519685039371 estimated_peak_memory_range: - min: 225280 - max: 1456568 + min: 233472 + max: 1465808 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 17 - job_id: jo5m9987g + job_id: jnp1mq47p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:06:44Z' + timestamp: '2024-08-27T00:22:06Z' - torchscript_onnx_tflite: - inference_time: 2269.0 - throughput: 440.72278536800354 + inference_time: 2144.0 + throughput: 466.4179104477612 estimated_peak_memory_range: min: 16384 - max: 23917264 + max: 22619344 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 17 - job_id: jz5wyy44g + job_id: jwgo9e2kg job_status: Passed torchscript_onnx_qnn: - inference_time: 1222.0 - throughput: 818.3306055646481 + inference_time: 1233.0 + throughput: 811.0300081103001 estimated_peak_memory_range: - min: 208896 - max: 13714640 + min: 212992 + max: 13972560 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 17 - job_id: jqpy88m0g + job_id: j0pxzekl5 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:06:49Z' + timestamp: '2024-08-27T00:22:10Z' - torchscript_onnx_tflite: - inference_time: 1401.0 - throughput: 713.7758743754462 + inference_time: 1419.0 + throughput: 704.7216349541931 estimated_peak_memory_range: - min: 24576 - max: 1456928 + min: 28672 + max: 9210736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 17 - job_id: jmg9oodmg + job_id: j1pvnz6rg job_status: Passed torchscript_onnx_qnn: - inference_time: 1005.0 - throughput: 995.0248756218906 + inference_time: 1010.0 + throughput: 990.0990099009902 estimated_peak_memory_range: min: 229376 - max: 1473232 + max: 1423752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 17 - job_id: jegn11kjp + job_id: jvgdm7xzg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:06:45Z' + timestamp: '2024-08-27T00:22:07Z' - torchscript_onnx_tflite: - inference_time: 1446.0 - throughput: 691.5629322268327 + inference_time: 1365.0 + throughput: 732.6007326007326 estimated_peak_memory_range: - min: 28672 - max: 1438168 + min: 36864 + max: 1727592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 17 - job_id: jnp1oo6n5 + job_id: j7gj8kve5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1008.0 - throughput: 992.063492063492 + inference_time: 1009.0 + throughput: 991.0802775024777 estimated_peak_memory_range: - min: 229376 - max: 1672888 + min: 233472 + max: 1591016 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 17 - job_id: joprxxwkp + job_id: jz578vy9p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:06:46Z' + timestamp: '2024-08-27T00:22:08Z' - torchscript_onnx_tflite: - inference_time: 1397.0 - throughput: 715.8196134574088 + inference_time: 1416.0 + throughput: 706.2146892655368 estimated_peak_memory_range: - min: 16384 - max: 1528088 + min: 9482240 + max: 10925536 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 17 - job_id: jvgd6626p + job_id: jlpen4dvp job_status: Passed torchscript_onnx_qnn: - inference_time: 1012.0 - throughput: 988.1422924901186 + inference_time: 993.0 + throughput: 1007.0493454179255 estimated_peak_memory_range: min: 229376 - max: 1754160 + max: 1509600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 17 - job_id: jep2ooe6g + job_id: jqp42jl1g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:06:47Z' + timestamp: '2024-08-27T00:22:09Z' - torchscript_onnx_qnn: - inference_time: 1105.0 - throughput: 904.9773755656108 + inference_time: 1296.0 + throughput: 771.604938271605 estimated_peak_memory_range: - min: 221184 - max: 221184 + min: 212992 + max: 212992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 17 - job_id: j0px00x8p + job_id: jmg9q9l8p job_status: Passed torchscript_onnx: - inference_time: 1659.0 - throughput: 602.7727546714889 + inference_time: 1540.0 + throughput: 649.3506493506494 estimated_peak_memory_range: - min: 8966144 - max: 8966144 + min: 8867840 + max: 8867840 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 19 - job_id: jogk664v5 + job_id: jopr71v7g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:06:52Z' + timestamp: '2024-08-27T00:22:13Z' diff --git a/qai_hub_models/models/quicksrnetmedium_quantized/perf.yaml b/qai_hub_models/models/quicksrnetmedium_quantized/perf.yaml index c9961910..e346d26c 100644 --- a/qai_hub_models/models/quicksrnetmedium_quantized/perf.yaml +++ b/qai_hub_models/models/quicksrnetmedium_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: QuickSRNetMedium-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 1168.0 - throughput: 856.1643835616438 + inference_time: 1165.0 + throughput: 858.3690987124463 estimated_peak_memory_range: - min: 28672 - max: 1418864 + min: 24576 + max: 79376296 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jz5wyy1mg + job_id: jn5qd7oog job_status: Passed torchscript_onnx_qnn: - inference_time: 517.0 - throughput: 1934.2359767891683 + inference_time: 512.0 + throughput: 1953.125 estimated_peak_memory_range: - min: 16384 - max: 2619320 + min: 0 + max: 68818296 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: joprxxn7p + job_id: jz5wrm6mp job_status: Passed torchscript_onnx: - inference_time: 745.0 - throughput: 1342.2818791946308 + inference_time: 764.0 + throughput: 1308.9005235602094 estimated_peak_memory_range: min: 16384 - max: 24404448 + max: 68046520 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 14 - job_id: j1p3oo6np + job_id: jopr7137g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:06:12Z' + timestamp: '2024-08-27T00:21:37Z' - torchscript_onnx_tflite: - inference_time: 935.0 - throughput: 1069.51871657754 + inference_time: 894.0 + throughput: 1118.5682326621925 estimated_peak_memory_range: - min: 12288 - max: 20684000 + min: 0 + max: 22264144 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jmg9oox8g + job_id: j1glqemmp job_status: Passed torchscript_onnx_qnn: - inference_time: 351.0 - throughput: 2849.002849002849 + inference_time: 357.0 + throughput: 2801.1204481792715 estimated_peak_memory_range: min: 65536 - max: 10768816 + max: 12368912 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: jep2oovqg + job_id: jmg9q9n8p job_status: Passed torchscript_onnx: - inference_time: 507.0 - throughput: 1972.3865877712033 + inference_time: 536.0 + throughput: 1865.6716417910447 estimated_peak_memory_range: min: 0 - max: 22402144 + max: 23073440 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 14 - job_id: jwgodd8k5 + job_id: jep2z3yqp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:06:13Z' + timestamp: '2024-08-27T00:21:38Z' - torchscript_onnx_tflite: - inference_time: 1178.0 - throughput: 848.8964346349745 + inference_time: 1121.0 + throughput: 892.0606601248885 estimated_peak_memory_range: - min: 28672 - max: 1288240 + min: 36864 + max: 14407080 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jnp1oov75 + job_id: jw560q4y5 job_status: Passed torchscript_onnx_qnn: - inference_time: 440.0 - throughput: 2272.7272727272725 + inference_time: 439.0 + throughput: 2277.904328018223 estimated_peak_memory_range: min: 73728 - max: 1278160 + max: 1298064 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: j2p0oovnp + job_id: jvgdm71zg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:06:05Z' + timestamp: '2024-08-27T00:21:31Z' - torchscript_onnx_tflite: - inference_time: 1449.0 - throughput: 690.1311249137336 + inference_time: 1382.0 + throughput: 723.589001447178 estimated_peak_memory_range: - min: 12288 - max: 23603872 + min: 20480 + max: 23414832 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jvgd66zzp + job_id: j1p3rq0np job_status: Passed torchscript_onnx_qnn: - inference_time: 590.0 - throughput: 1694.915254237288 + inference_time: 581.0 + throughput: 1721.170395869191 estimated_peak_memory_range: - min: 61440 - max: 11291920 + min: 65536 + max: 13262080 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: j1glwwlmp + job_id: jo5mlvx9g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:06:10Z' + timestamp: '2024-08-27T00:21:36Z' - torchscript_onnx_tflite: - inference_time: 1160.0 - throughput: 862.0689655172414 + inference_time: 1113.0 + throughput: 898.4725965858041 estimated_peak_memory_range: - min: 16384 - max: 1438456 + min: 28672 + max: 1404896 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jz57oo79g + job_id: jwgo9e6kg job_status: Passed torchscript_onnx_qnn: - inference_time: 452.0 - throughput: 2212.3893805309735 + inference_time: 445.0 + throughput: 2247.191011235955 estimated_peak_memory_range: - min: 73728 - max: 1443328 + min: 81920 + max: 1468312 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: j1p8jj4o5 + job_id: jz578vr9p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:06:06Z' + timestamp: '2024-08-27T00:21:32Z' - torchscript_onnx_tflite: - inference_time: 1167.0 - throughput: 856.898029134533 + inference_time: 1113.0 + throughput: 898.4725965858041 estimated_peak_memory_range: - min: 24576 - max: 1359128 + min: 69632 + max: 12499856 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jqp4ee91g + job_id: j1pvnzkrg job_status: Passed torchscript_onnx_qnn: - inference_time: 475.0 - throughput: 2105.2631578947367 + inference_time: 447.0 + throughput: 2237.136465324385 estimated_peak_memory_range: - min: 81920 - max: 1499168 + min: 86016 + max: 1701272 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: jogk669n5 + job_id: jqp42jr1g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:06:07Z' + timestamp: '2024-08-27T00:21:33Z' - torchscript_onnx_tflite: - inference_time: 1172.0 - throughput: 853.2423208191126 + inference_time: 1110.0 + throughput: 900.9009009009009 estimated_peak_memory_range: - min: 12288 - max: 1351120 + min: 28672 + max: 1514240 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: j0px00dlp + job_id: j7gj8kne5 job_status: Passed torchscript_onnx_qnn: - inference_time: 456.0 - throughput: 2192.9824561403507 + inference_time: 440.0 + throughput: 2272.7272727272725 estimated_peak_memory_range: - min: 81920 - max: 1273224 + min: 28672 + max: 1815944 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: jn5q44mog + job_id: j0pxzeol5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:06:08Z' + timestamp: '2024-08-27T00:21:34Z' - torchscript_onnx_tflite: - inference_time: 2594.0 - throughput: 385.50501156515037 + inference_time: 2553.0 + throughput: 391.6960438699569 estimated_peak_memory_range: - min: 3194880 - max: 18825184 + min: 3203072 + max: 19319584 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jo5m99d9g + job_id: jlpen4mvp job_status: Passed torchscript_onnx_qnn: - inference_time: 1084.0 - throughput: 922.509225092251 + inference_time: 1053.0 + throughput: 949.667616334283 estimated_peak_memory_range: - min: 81920 - max: 8193264 + min: 12288 + max: 7832832 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: jw56oowy5 + job_id: jegnwrvqg job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:06:11Z' + timestamp: '2024-08-27T00:21:37Z' - torchscript_onnx_tflite: - inference_time: 13005.0 - throughput: 76.89350249903883 + inference_time: 11756.0 + throughput: 85.06294658046954 estimated_peak_memory_range: - min: 3366912 - max: 6553816 + min: 3457024 + max: 11721552 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jegn117qp + job_id: jygz0vdx5 job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:06:00Z' + timestamp: '2024-08-27T00:21:26Z' - torchscript_onnx_qnn: - inference_time: 552.0 - throughput: 1811.5942028985507 + inference_time: 546.0 + throughput: 1831.5018315018315 estimated_peak_memory_range: - min: 61440 - max: 61440 + min: 57344 + max: 57344 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 10 - job_id: jqpy887lg + job_id: jnp1mqz7p job_status: Passed torchscript_onnx: - inference_time: 776.0 - throughput: 1288.659793814433 + inference_time: 777.0 + throughput: 1287.001287001287 estimated_peak_memory_range: - min: 3403776 - max: 3403776 + min: 3395584 + max: 3395584 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 14 - job_id: j1pv227rg + job_id: jqpyyv3lp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:06:14Z' + timestamp: '2024-08-27T00:21:39Z' diff --git a/qai_hub_models/models/quicksrnetsmall/perf.yaml b/qai_hub_models/models/quicksrnetsmall/perf.yaml index 5198ece0..2706b47f 100644 --- a/qai_hub_models/models/quicksrnetsmall/perf.yaml +++ b/qai_hub_models/models/quicksrnetsmall/perf.yaml @@ -45,11 +45,11 @@ models: - name: QuickSRNetSmall performance_metrics: - torchscript_onnx_tflite: - inference_time: 1388.0 - throughput: 720.4610951008646 + inference_time: 1352.0 + throughput: 739.6449704142012 estimated_peak_memory_range: min: 24576 - max: 8499600 + max: 17929040 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 11 - job_id: jnp1oo775 + job_id: j7gj804v5 job_status: Passed torchscript_onnx_qnn: - inference_time: 997.0 - throughput: 1003.0090270812437 + inference_time: 1018.0 + throughput: 982.3182711198428 estimated_peak_memory_range: - min: 212992 - max: 2381680 + min: 16384 + max: 62959688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 11 - job_id: joprxxo7p + job_id: jz5wrdnmp job_status: Passed torchscript_onnx: - inference_time: 1482.0 - throughput: 674.7638326585695 + inference_time: 1483.0 + throughput: 674.3088334457182 estimated_peak_memory_range: min: 212992 - max: 1888320 + max: 8712984 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 13 - job_id: jw56oody5 + job_id: jegnw98qg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:05:30Z' + timestamp: '2024-08-27T00:20:59Z' - torchscript_onnx_tflite: - inference_time: 835.0 - throughput: 1197.6047904191616 + inference_time: 816.0 + throughput: 1225.4901960784314 estimated_peak_memory_range: - min: 0 - max: 20537280 + min: 16384 + max: 20788048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 11 - job_id: jvgd668zp + job_id: jlpenr3op job_status: Passed torchscript_onnx_qnn: inference_time: 626.0 throughput: 1597.444089456869 estimated_peak_memory_range: - min: 0 - max: 10746848 + min: 208896 + max: 11613824 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 11 - job_id: jep2oo4qg + job_id: jmg9q3e8p job_status: Passed torchscript_onnx: - inference_time: 966.0 - throughput: 1035.1966873706003 + inference_time: 975.0 + throughput: 1025.6410256410256 estimated_peak_memory_range: min: 0 - max: 21113552 + max: 21092352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 13 - job_id: j1p3oownp + job_id: jopr74j7g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:05:31Z' + timestamp: '2024-08-27T00:21:00Z' - torchscript_onnx_tflite: - inference_time: 1392.0 - throughput: 718.3908045977012 + inference_time: 1349.0 + throughput: 741.2898443291327 estimated_peak_memory_range: - min: 28672 - max: 1376792 + min: 24576 + max: 1317080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 11 - job_id: jz57ook9g + job_id: jygz0xko5 job_status: Passed torchscript_onnx_qnn: - inference_time: 983.0 - throughput: 1017.293997965412 + inference_time: 971.0 + throughput: 1029.8661174047375 estimated_peak_memory_range: - min: 229376 - max: 1540616 + min: 16384 + max: 4020256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 11 - job_id: j2p0oodnp + job_id: jvgdmrlzg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:05:24Z' + timestamp: '2024-08-27T00:20:54Z' - torchscript_onnx_tflite: - inference_time: 1906.0 - throughput: 524.6589716684156 + inference_time: 2274.0 + throughput: 439.7537379067722 estimated_peak_memory_range: - min: 16384 - max: 21539456 + min: 12607488 + max: 32723776 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 11 - job_id: jqp4eem1g + job_id: jz5wrdn3p job_status: Passed torchscript_onnx_qnn: - inference_time: 1119.0 - throughput: 893.6550491510277 + inference_time: 1114.0 + throughput: 897.6660682226212 estimated_peak_memory_range: - min: 45056 - max: 12566800 + min: 49152 + max: 12541168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 11 - job_id: j1glww1mp + job_id: jo5mlwy9g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:05:29Z' + timestamp: '2024-08-27T00:20:58Z' - torchscript_onnx_tflite: - inference_time: 1432.0 - throughput: 698.3240223463687 + inference_time: 1317.0 + throughput: 759.3014426727411 estimated_peak_memory_range: min: 24576 - max: 1529824 + max: 31364488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 11 - job_id: j0px003lp + job_id: jmg9q3ewp job_status: Passed torchscript_onnx_qnn: - inference_time: 981.0 - throughput: 1019.367991845056 + inference_time: 1087.0 + throughput: 919.9632014719411 estimated_peak_memory_range: - min: 225280 - max: 1308856 + min: 32768 + max: 4122176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 11 - job_id: j1p8jj6o5 + job_id: jz578j39p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:05:25Z' + timestamp: '2024-08-27T00:20:55Z' - torchscript_onnx_tflite: - inference_time: 1375.0 - throughput: 727.2727272727273 + inference_time: 2591.0 + throughput: 385.95137012736393 estimated_peak_memory_range: - min: 28672 - max: 1519048 + min: 9478144 + max: 19902360 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 11 - job_id: jo5m99o9g + job_id: jnp1mdx8p job_status: Passed torchscript_onnx_qnn: - inference_time: 990.0 - throughput: 1010.10101010101 + inference_time: 981.0 + throughput: 1019.367991845056 estimated_peak_memory_range: min: 229376 - max: 1740624 + max: 1587248 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 11 - job_id: jogk66on5 + job_id: jqp42x01g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:05:27Z' + timestamp: '2024-08-27T00:20:56Z' - torchscript_onnx_tflite: - inference_time: 1391.0 - throughput: 718.9072609633357 + inference_time: 1375.0 + throughput: 727.2727272727273 estimated_peak_memory_range: - min: 24576 - max: 1348736 + min: 28672 + max: 1379064 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 11 - job_id: jegn11oqp + job_id: jvgdmrlrg job_status: Passed torchscript_onnx_qnn: - inference_time: 991.0 - throughput: 1009.0817356205853 + inference_time: 1001.0 + throughput: 999.000999000999 estimated_peak_memory_range: - min: 229376 - max: 1564840 + min: 12288 + max: 7107168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 11 - job_id: jn5q44zog + job_id: j0pxz72l5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:05:28Z' + timestamp: '2024-08-27T00:20:57Z' - torchscript_onnx_qnn: - inference_time: 1104.0 - throughput: 905.7971014492754 + inference_time: 1087.0 + throughput: 919.9632014719411 estimated_peak_memory_range: - min: 208896 - max: 208896 + min: 204800 + max: 204800 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 11 - job_id: jqpy88qlg + job_id: jnp1mdx7p job_status: Passed torchscript_onnx: - inference_time: 1463.0 - throughput: 683.526999316473 + inference_time: 1464.0 + throughput: 683.0601092896175 estimated_peak_memory_range: - min: 8978432 - max: 8978432 + min: 8998912 + max: 8998912 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 13 - job_id: jwgodd4k5 + job_id: jep2z7nqp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:05:32Z' + timestamp: '2024-08-27T00:21:01Z' diff --git a/qai_hub_models/models/quicksrnetsmall_quantized/perf.yaml b/qai_hub_models/models/quicksrnetsmall_quantized/perf.yaml index dfd00bcc..3fd1ed4e 100644 --- a/qai_hub_models/models/quicksrnetsmall_quantized/perf.yaml +++ b/qai_hub_models/models/quicksrnetsmall_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: QuickSRNetSmall-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 1122.0 - throughput: 891.2655971479501 + inference_time: 1129.0 + throughput: 885.7395925597874 estimated_peak_memory_range: - min: 28672 - max: 77084656 + min: 3956736 + max: 5500928 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: jmg9oo88g + job_id: j1pvn1okg job_status: Passed torchscript_onnx_qnn: - inference_time: 469.0 - throughput: 2132.1961620469083 + inference_time: 473.0 + throughput: 2114.164904862579 estimated_peak_memory_range: - min: 12288 - max: 20129424 + min: 69632 + max: 2290256 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: jep2oojqg + job_id: jqp42xo8g job_status: Passed torchscript_onnx: - inference_time: 729.0 - throughput: 1371.7421124828531 + inference_time: 700.0 + throughput: 1428.5714285714287 estimated_peak_memory_range: min: 69632 - max: 12098632 + max: 12266536 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 12 - job_id: jwgoddok5 + job_id: jogkklnwg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:04:53Z' + timestamp: '2024-08-27T00:20:25Z' - torchscript_onnx_tflite: - inference_time: 886.0 - throughput: 1128.6681715575621 + inference_time: 871.0 + throughput: 1148.105625717566 estimated_peak_memory_range: min: 0 - max: 19945216 + max: 20475248 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: jnp1oo375 + job_id: j7gj80mv5 job_status: Passed torchscript_onnx_qnn: - inference_time: 329.0 - throughput: 3039.51367781155 + inference_time: 315.0 + throughput: 3174.6031746031745 estimated_peak_memory_range: - min: 69632 - max: 10954720 + min: 65536 + max: 12198496 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: jqpy88nlg + job_id: j0pxz7j35 job_status: Passed torchscript_onnx: - inference_time: 516.0 - throughput: 1937.984496124031 + inference_time: 508.0 + throughput: 1968.5039370078741 estimated_peak_memory_range: min: 0 - max: 20504544 + max: 20977760 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 12 - job_id: j1pv22erg + job_id: jn5qd7kng job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:04:54Z' + timestamp: '2024-08-27T00:20:26Z' - torchscript_onnx_tflite: - inference_time: 1191.0 - throughput: 839.6305625524769 + inference_time: 1073.0 + throughput: 931.9664492078285 estimated_peak_memory_range: - min: 811008 - max: 2310016 + min: 12288 + max: 1320168 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: jvgd660zp + job_id: jlpenr1op job_status: Passed torchscript_onnx_qnn: - inference_time: 449.0 - throughput: 2227.1714922048996 + inference_time: 450.0 + throughput: 2222.222222222222 estimated_peak_memory_range: - min: 90112 - max: 1290776 + min: 73728 + max: 1289232 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: j1p8jj8o5 + job_id: jegnw9ykg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:04:45Z' + timestamp: '2024-08-27T00:20:20Z' - torchscript_onnx_tflite: - inference_time: 2512.0 - throughput: 398.0891719745223 + inference_time: 2070.0 + throughput: 483.09178743961354 estimated_peak_memory_range: min: 12288 - max: 21130912 + max: 21393232 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: jz57oo69g + job_id: jygz0x9o5 job_status: Passed torchscript_onnx_qnn: - inference_time: 538.0 - throughput: 1858.736059479554 + inference_time: 540.0 + throughput: 1851.851851851852 estimated_peak_memory_range: min: 65536 - max: 11678880 + max: 13131376 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: jw56oory5 + job_id: j2p0x1q9p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:04:50Z' + timestamp: '2024-08-27T00:20:24Z' - torchscript_onnx_tflite: - inference_time: 1182.0 - throughput: 846.0236886632825 + inference_time: 1068.0 + throughput: 936.3295880149813 estimated_peak_memory_range: - min: 32768 - max: 77011192 + min: 20480 + max: 14109312 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: jqp4ee81g + job_id: jz5wrdv3p job_status: Passed torchscript_onnx_qnn: - inference_time: 451.0 - throughput: 2217.2949002217297 + inference_time: 450.0 + throughput: 2222.222222222222 estimated_peak_memory_range: - min: 73728 - max: 1420536 + min: 69632 + max: 1337624 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: jogk66dn5 + job_id: jopr74q0g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:04:46Z' + timestamp: '2024-08-27T00:20:21Z' - torchscript_onnx_tflite: - inference_time: 1111.0 - throughput: 900.0900090009001 + inference_time: 1053.0 + throughput: 949.667616334283 estimated_peak_memory_range: min: 24576 - max: 1467064 + max: 1377568 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: j0px00mlp + job_id: jmg9q31wp job_status: Passed torchscript_onnx_qnn: - inference_time: 451.0 - throughput: 2217.2949002217297 + inference_time: 453.0 + throughput: 2207.5055187637968 estimated_peak_memory_range: - min: 77824 - max: 2263592 + min: 102400 + max: 1310984 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: jn5q44wog + job_id: jep2z76rp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:04:48Z' + timestamp: '2024-08-27T00:20:22Z' - torchscript_onnx_tflite: - inference_time: 1122.0 - throughput: 891.2655971479501 + inference_time: 1062.0 + throughput: 941.6195856873823 estimated_peak_memory_range: - min: 20480 - max: 1619008 + min: 16384 + max: 2950976 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: jo5m9949g + job_id: jnp1mdl8p job_status: Passed torchscript_onnx_qnn: - inference_time: 459.0 - throughput: 2178.649237472767 + inference_time: 455.0 + throughput: 2197.802197802198 estimated_peak_memory_range: - min: 20480 - max: 1498600 + min: 86016 + max: 1293008 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: j1glwwomp + job_id: jqpyy4w8p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:04:49Z' + timestamp: '2024-08-27T00:20:23Z' - torchscript_onnx_tflite: - inference_time: 2459.0 - throughput: 406.669377795852 + inference_time: 2269.0 + throughput: 440.72278536800354 estimated_peak_memory_range: - min: 12288 - max: 15356512 + min: 3190784 + max: 18560256 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: jegn11xqp + job_id: jvgdmr9rg job_status: Passed torchscript_onnx_qnn: - inference_time: 958.0 - throughput: 1043.8413361169103 + inference_time: 951.0 + throughput: 1051.5247108307046 estimated_peak_memory_range: - min: 28672 - max: 8406096 + min: 65536 + max: 7982784 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: j1p3ooxnp + job_id: j1p8k39kp job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:04:52Z' + timestamp: '2024-08-27T00:20:25Z' - torchscript_onnx_tflite: - inference_time: 11640.0 - throughput: 85.91065292096219 + inference_time: 9581.0 + throughput: 104.3732387015969 estimated_peak_memory_range: - min: 3375104 - max: 10435440 + min: 3313664 + max: 8865144 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 13 - job_id: joprxx97p + job_id: jz578jwvp job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:04:40Z' + timestamp: '2024-08-27T00:20:14Z' - torchscript_onnx_qnn: - inference_time: 559.0 - throughput: 1788.9087656529516 + inference_time: 555.0 + throughput: 1801.8018018018017 estimated_peak_memory_range: - min: 65536 - max: 65536 + min: 61440 + max: 61440 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 7 - job_id: j2p0ooknp + job_id: jo5mlw2dg job_status: Passed torchscript_onnx: - inference_time: 730.0 - throughput: 1369.86301369863 + inference_time: 736.0 + throughput: 1358.695652173913 estimated_peak_memory_range: - min: 3358720 - max: 3358720 + min: 3391488 + max: 3391488 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 12 - job_id: j7gj33oep + job_id: j1glq0zjp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:04:55Z' + timestamp: '2024-08-27T00:20:27Z' diff --git a/qai_hub_models/models/real_esrgan_general_x4v3/perf.yaml b/qai_hub_models/models/real_esrgan_general_x4v3/perf.yaml index 3c88bed0..3f5c021f 100644 --- a/qai_hub_models/models/real_esrgan_general_x4v3/perf.yaml +++ b/qai_hub_models/models/real_esrgan_general_x4v3/perf.yaml @@ -45,11 +45,11 @@ models: - name: Real-ESRGAN-General-x4v3 performance_metrics: - torchscript_onnx_tflite: - inference_time: 7314.0 - throughput: 136.7240907847963 + inference_time: 7328.0 + throughput: 136.46288209606988 estimated_peak_memory_range: - min: 16617472 - max: 26134192 + min: 15810560 + max: 17426576 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 72 - job_id: jvgd667zp + job_id: jlpenr9op job_status: Passed torchscript_onnx_qnn: - inference_time: 6263.0 - throughput: 159.6678907871627 + inference_time: 6290.0 + throughput: 158.9825119236884 estimated_peak_memory_range: - min: 20480 - max: 8461112 + min: 118784 + max: 8626696 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jep2oo3qg + job_id: jqp42xq8g job_status: Passed torchscript_onnx: - inference_time: 6898.0 - throughput: 144.96955639315743 + inference_time: 6946.0 + throughput: 143.96775122372588 estimated_peak_memory_range: - min: 9371648 - max: 13584560 + min: 6365184 + max: 10851576 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: j1p3oo8np + job_id: j1p8k3okp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:04:11Z' + timestamp: '2024-08-27T00:19:48Z' - torchscript_onnx_tflite: - inference_time: 5421.0 - throughput: 184.46781036709095 + inference_time: 5358.0 + throughput: 186.6368047779022 estimated_peak_memory_range: - min: 10604544 - max: 70358064 + min: 11898880 + max: 72127520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 72 - job_id: jz57oov9g + job_id: jygz0xeo5 job_status: Passed torchscript_onnx_qnn: - inference_time: 4612.0 - throughput: 216.8256721595837 + inference_time: 4587.0 + throughput: 218.00741225201656 estimated_peak_memory_range: - min: 204800 - max: 18687728 + min: 0 + max: 19184128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jqpy88vlg + job_id: j0pxz7v35 job_status: Passed torchscript_onnx: - inference_time: 5149.0 - throughput: 194.21246844047388 + inference_time: 5144.0 + throughput: 194.4012441679627 estimated_peak_memory_range: - min: 6516736 - max: 71913008 + min: 6832128 + max: 72953536 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: jwgoddmk5 + job_id: jogkklzwg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:04:12Z' + timestamp: '2024-08-27T00:19:49Z' - torchscript_onnx_tflite: - inference_time: 7457.0 - throughput: 134.10218586562962 + inference_time: 7348.0 + throughput: 136.09145345672292 estimated_peak_memory_range: - min: 15806464 - max: 17164888 + min: 15781888 + max: 135272472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 72 - job_id: jqp4eej1g + job_id: jz5wrdo3p job_status: Passed torchscript_onnx_qnn: - inference_time: 5733.0 - throughput: 174.4287458573173 + inference_time: 5739.0 + throughput: 174.24638438752396 estimated_peak_memory_range: - min: 16384 - max: 3865176 + min: 221184 + max: 1494368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: j1p8jjwo5 + job_id: jegnw92kg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:04:05Z' + timestamp: '2024-08-27T00:19:43Z' - torchscript_onnx_tflite: - inference_time: 10743.0 - throughput: 93.08386856557759 + inference_time: 10844.0 + throughput: 92.21689413500553 estimated_peak_memory_range: - min: 11517952 - max: 75538064 + min: 11538432 + max: 75616368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 72 - job_id: j0px00elp + job_id: jmg9q3vwp job_status: Passed torchscript_onnx_qnn: - inference_time: 9704.0 - throughput: 103.05028854080791 + inference_time: 9585.0 + throughput: 104.32968179447053 estimated_peak_memory_range: - min: 196608 - max: 23533808 + min: 208896 + max: 26361584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jw56oovy5 + job_id: j2p0x1y9p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:04:10Z' + timestamp: '2024-08-27T00:19:47Z' - torchscript_onnx_tflite: - inference_time: 7394.0 - throughput: 135.2447930754666 + inference_time: 7242.0 + throughput: 138.08340237503452 estimated_peak_memory_range: - min: 15761408 - max: 17305560 + min: 15794176 + max: 135347496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 72 - job_id: jo5m99v9g + job_id: jnp1md08p job_status: Passed torchscript_onnx_qnn: - inference_time: 5749.0 - throughput: 173.94329448599757 + inference_time: 5807.0 + throughput: 172.20595832615808 estimated_peak_memory_range: - min: 270336 - max: 1713200 + min: 229376 + max: 1733528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jogk66rn5 + job_id: jopr74k0g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:04:06Z' + timestamp: '2024-08-27T00:19:44Z' - torchscript_onnx_tflite: - inference_time: 7398.0 - throughput: 135.17166801838334 + inference_time: 7286.0 + throughput: 137.24951962668132 estimated_peak_memory_range: - min: 15773696 - max: 20186928 + min: 15777792 + max: 19819600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 72 - job_id: jegn11rqp + job_id: jvgdmrwrg job_status: Passed torchscript_onnx_qnn: - inference_time: 5765.0 - throughput: 173.46053772766695 + inference_time: 5819.0 + throughput: 171.85083347654236 estimated_peak_memory_range: - min: 249856 - max: 1494720 + min: 253952 + max: 1801872 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: jn5q449og + job_id: jep2z78rp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:04:07Z' + timestamp: '2024-08-27T00:19:45Z' - torchscript_onnx_tflite: - inference_time: 7443.0 - throughput: 134.35442697836893 + inference_time: 7323.0 + throughput: 136.5560562610952 estimated_peak_memory_range: - min: 16150528 - max: 17646416 + min: 15749120 + max: 17132656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 72 - job_id: joprxx17p + job_id: jz578jzvp job_status: Passed torchscript_onnx_qnn: - inference_time: 5758.0 - throughput: 173.67141368530739 + inference_time: 5809.0 + throughput: 172.14666896195558 estimated_peak_memory_range: - min: 270336 - max: 1486264 + min: 61440 + max: 4067456 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: j1glww7mp + job_id: jqpyy4e8p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:04:08Z' + timestamp: '2024-08-27T00:19:46Z' - torchscript_onnx_qnn: - inference_time: 6141.0 - throughput: 162.83992835043153 + inference_time: 6349.0 + throughput: 157.50511891636478 estimated_peak_memory_range: min: 212992 max: 212992 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 72 - job_id: j2p0ooenp + job_id: jo5mlwrdg job_status: Passed torchscript_onnx: - inference_time: 7052.0 - throughput: 141.80374361883153 + inference_time: 7058.0 + throughput: 141.68319637291017 estimated_peak_memory_range: - min: 8855552 - max: 8855552 + min: 8994816 + max: 8994816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 74 - job_id: j1pv224rg + job_id: jn5qd78ng job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:04:13Z' + timestamp: '2024-08-27T00:19:50Z' diff --git a/qai_hub_models/models/real_esrgan_x4plus/perf.yaml b/qai_hub_models/models/real_esrgan_x4plus/perf.yaml index 1d9d80ed..0a5df3d7 100644 --- a/qai_hub_models/models/real_esrgan_x4plus/perf.yaml +++ b/qai_hub_models/models/real_esrgan_x4plus/perf.yaml @@ -45,11 +45,11 @@ models: - name: Real-ESRGAN-x4plus performance_metrics: - torchscript_onnx_tflite: - inference_time: 65101.0 - throughput: 15.360747146741218 + inference_time: 65006.0 + throughput: 15.383195397347937 estimated_peak_memory_range: - min: 4050944 - max: 6100024 + min: 4313088 + max: 7548280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: joprxx40p + job_id: jygz0x1o5 job_status: Passed torchscript_onnx_qnn: - inference_time: 70426.0 - throughput: 14.199301394371396 + inference_time: 68782.0 + throughput: 14.538687447297258 estimated_peak_memory_range: - min: 167936 - max: 101926528 + min: 0 + max: 103325840 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1029 - job_id: j1glww0jp + job_id: j0pxz7w35 job_status: Passed torchscript_onnx: - inference_time: 67700.0 - throughput: 14.771048744460856 + inference_time: 72197.0 + throughput: 13.850991038408798 estimated_peak_memory_range: - min: 6782976 - max: 9519544 + min: 16384 + max: 44568080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1030 - job_id: jz5wyym3g + job_id: jn5qd7jng job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:03:32Z' + timestamp: '2024-08-27T00:19:13Z' - torchscript_onnx_tflite: - inference_time: 53016.0 - throughput: 18.862230270107137 + inference_time: 50182.0 + throughput: 19.927464030927425 estimated_peak_memory_range: - min: 3244032 - max: 610975632 + min: 16384 + max: 608203072 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: jep2oo7rg + job_id: jz5wrdj3p job_status: Passed torchscript_onnx_qnn: - inference_time: 50418.0 - throughput: 19.834186203340078 + inference_time: 51219.0 + throughput: 19.524004763857164 estimated_peak_memory_range: - min: 26656768 - max: 130744128 + min: 32768 + max: 103147920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1029 - job_id: jw56oo365 + job_id: jo5mlwjdg job_status: Passed torchscript_onnx: - inference_time: 52104.0 - throughput: 19.19238446184554 + inference_time: 51902.0 + throughput: 19.26708026665639 estimated_peak_memory_range: - min: 8863744 - max: 636986464 + min: 6889472 + max: 633862624 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1030 - job_id: jmg9oo9wg + job_id: j1glq0jjp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:03:33Z' + timestamp: '2024-08-27T00:19:14Z' - torchscript_onnx_tflite: - inference_time: 63842.0 - throughput: 15.663669684533692 + inference_time: 63418.0 + throughput: 15.768393831404333 estimated_peak_memory_range: - min: 3211264 - max: 7358312 + min: 3198976 + max: 7102544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: jqpy8848g + job_id: jmg9q36wp job_status: Passed torchscript_onnx_qnn: - inference_time: 64259.0 - throughput: 15.562022440436358 + inference_time: 63420.0 + throughput: 15.76789656259855 estimated_peak_memory_range: - min: 184320 - max: 1524392 + min: 348160 + max: 1610928 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1029 - job_id: jwgodd1q5 + job_id: jopr74z0g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:03:25Z' + timestamp: '2024-08-27T00:19:08Z' - torchscript_onnx_tflite: - inference_time: 134044.0 - throughput: 7.460236937125123 + inference_time: 151457.0 + throughput: 6.602534052569376 estimated_peak_memory_range: - min: 3559424 - max: 576459856 + min: 3477504 + max: 576961152 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: j2p0oo19p + job_id: jnp1mdr8p job_status: Passed torchscript_onnx_qnn: - inference_time: 126094.0 - throughput: 7.930591463511349 + inference_time: 129895.0 + throughput: 7.698525732322261 estimated_peak_memory_range: - min: 315392 - max: 78117360 + min: 307200 + max: 76255984 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1029 - job_id: jygzzzvog + job_id: jogkkljwg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:03:30Z' + timestamp: '2024-08-27T00:19:12Z' - torchscript_onnx_tflite: - inference_time: 65636.0 - throughput: 15.235541471143884 + inference_time: 68425.0 + throughput: 14.614541468761418 estimated_peak_memory_range: - min: 3457024 - max: 7412176 + min: 3248128 + max: 11204832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: j1p8jj3k5 + job_id: jvgdmrjrg job_status: Passed torchscript_onnx_qnn: - inference_time: 65860.0 - throughput: 15.183723048891588 + inference_time: 63500.0 + throughput: 15.748031496062993 estimated_peak_memory_range: - min: 0 - max: 4045520 + min: 421888 + max: 1716896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1029 - job_id: j1pv22zkg + job_id: jep2z72rp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:03:26Z' + timestamp: '2024-08-27T00:19:09Z' - torchscript_onnx_tflite: - inference_time: 63215.0 - throughput: 15.819030293443012 + inference_time: 68220.0 + throughput: 14.65845793022574 estimated_peak_memory_range: - min: 3325952 - max: 6206136 + min: 3182592 + max: 7448608 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: jogk66lw5 + job_id: jz578jqvp job_status: Passed torchscript_onnx_qnn: - inference_time: 63667.0 - throughput: 15.70672404856519 + inference_time: 63795.0 + throughput: 15.675209655929148 estimated_peak_memory_range: - min: 368640 - max: 1635448 + min: 393216 + max: 1765024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1029 - job_id: j7gj33kvp + job_id: j2p0x1n9p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:03:28Z' + timestamp: '2024-08-27T00:19:10Z' - torchscript_onnx_tflite: - inference_time: 68890.0 - throughput: 14.515894904920888 + inference_time: 73925.0 + throughput: 13.527223537368956 estimated_peak_memory_range: - min: 3424256 - max: 6986096 + min: 356352 + max: 4191816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1028 - job_id: jn5q447ng + job_id: jqp42xz8g job_status: Passed torchscript_onnx_qnn: - inference_time: 63519.0 - throughput: 15.743320896109825 + inference_time: 62573.0 + throughput: 15.981333802119124 estimated_peak_memory_range: - min: 413696 - max: 1619992 + min: 372736 + max: 1783440 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1029 - job_id: jlpe664og + job_id: j1p8k3lkp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:03:29Z' + timestamp: '2024-08-27T00:19:11Z' - torchscript_onnx_qnn: - inference_time: 65451.0 - throughput: 15.278605368901927 + inference_time: 65118.0 + throughput: 15.356737000522129 estimated_peak_memory_range: - min: 208896 - max: 208896 + min: 217088 + max: 217088 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1029 - job_id: j1p3oo43p + job_id: jegnw9jkg job_status: Passed torchscript_onnx: - inference_time: 65745.0 - throughput: 15.210282150733896 + inference_time: 65471.0 + throughput: 15.273938079455025 estimated_peak_memory_range: - min: 40316928 - max: 40316928 + min: 40824832 + max: 40824832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1030 - job_id: jnp1ooq85 + job_id: jw5603k65 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:03:34Z' + timestamp: '2024-08-27T00:19:15Z' diff --git a/qai_hub_models/models/regnet/perf.yaml b/qai_hub_models/models/regnet/perf.yaml index 85872b7f..0136bffd 100644 --- a/qai_hub_models/models/regnet/perf.yaml +++ b/qai_hub_models/models/regnet/perf.yaml @@ -45,11 +45,11 @@ models: - name: RegNet performance_metrics: - torchscript_onnx_tflite: - inference_time: 2021.0 - throughput: 494.80455220188026 + inference_time: 2070.0 + throughput: 483.09178743961354 estimated_peak_memory_range: - min: 53248 - max: 2098528 + min: 12288 + max: 4836856 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 114 - job_id: jqpy88d8g + job_id: jz5wrdk3p job_status: Passed torchscript_onnx_qnn: inference_time: 2129.0 throughput: 469.7040864255519 estimated_peak_memory_range: - min: 12288 - max: 64558848 + min: 16384 + max: 64499744 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 188 - job_id: j1p3oom3p + job_id: jo5mlw7dg job_status: Passed torchscript_onnx: - inference_time: 2224.0 - throughput: 449.64028776978415 + inference_time: 2193.0 + throughput: 455.99635202918375 estimated_peak_memory_range: min: 12288 - max: 43907600 + max: 499222848 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 190 - job_id: jnp1ood85 + job_id: jn5qd7nng job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:02:36Z' + timestamp: '2024-08-27T00:18:23Z' - torchscript_onnx_tflite: - inference_time: 1409.0 - throughput: 709.7232079488999 + inference_time: 1416.0 + throughput: 706.2146892655368 estimated_peak_memory_range: - min: 12288 - max: 144517744 + min: 16384 + max: 144766256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 114 - job_id: j2p0oor9p + job_id: jmg9q3rwp job_status: Passed torchscript_onnx_qnn: - inference_time: 1452.0 - throughput: 688.7052341597796 + inference_time: 1475.0 + throughput: 677.9661016949152 estimated_peak_memory_range: min: 618496 - max: 30508672 + max: 28249456 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 188 - job_id: jwgoddvq5 + job_id: jegnw94kg job_status: Passed torchscript_onnx: - inference_time: 1574.0 - throughput: 635.3240152477764 + inference_time: 1563.0 + throughput: 639.7952655150352 estimated_peak_memory_range: min: 0 - max: 144860640 + max: 145638944 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 190 - job_id: jvgd66rrp + job_id: j1glq0djp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:02:37Z' + timestamp: '2024-08-27T00:18:24Z' - torchscript_onnx_tflite: - inference_time: 2010.0 - throughput: 497.5124378109453 + inference_time: 2004.0 + throughput: 499.001996007984 estimated_peak_memory_range: - min: 24576 - max: 1867800 + min: 49152 + max: 2154976 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 114 - job_id: j1p8jj7k5 + job_id: jnp1md98p job_status: Passed torchscript_onnx_qnn: - inference_time: 2030.0 - throughput: 492.61083743842363 + inference_time: 1990.0 + throughput: 502.51256281407035 estimated_peak_memory_range: - min: 634880 - max: 1969024 + min: 630784 + max: 1942416 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 188 - job_id: j7gj330vp + job_id: jep2z71rp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:02:30Z' + timestamp: '2024-08-27T00:18:18Z' - torchscript_onnx_tflite: - inference_time: 2831.0 - throughput: 353.2320734722713 + inference_time: 2795.0 + throughput: 357.78175313059035 estimated_peak_memory_range: - min: 28672 - max: 125691056 + min: 16384 + max: 129359056 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 114 - job_id: jogk66yw5 + job_id: jvgdmrkrg job_status: Passed torchscript_onnx_qnn: - inference_time: 2939.0 - throughput: 340.25178632187817 + inference_time: 2944.0 + throughput: 339.67391304347825 estimated_peak_memory_range: min: 618496 - max: 24374576 + max: 25017696 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 188 - job_id: jmg9oo3wg + job_id: jogkkl1wg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:02:35Z' + timestamp: '2024-08-27T00:18:22Z' - torchscript_onnx_tflite: - inference_time: 2018.0 - throughput: 495.5401387512388 + inference_time: 2010.0 + throughput: 497.5124378109453 estimated_peak_memory_range: - min: 16384 - max: 2121288 + min: 28672 + max: 2142816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 114 - job_id: jn5q442ng + job_id: jz578jmvp job_status: Passed torchscript_onnx_qnn: - inference_time: 2015.0 - throughput: 496.27791563275434 + inference_time: 1996.0 + throughput: 501.00200400801606 estimated_peak_memory_range: - min: 647168 - max: 2001480 + min: 630784 + max: 1917752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 188 - job_id: jlpe66rog + job_id: jqpyy4l8p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:02:31Z' + timestamp: '2024-08-27T00:18:19Z' - torchscript_onnx_tflite: - inference_time: 2063.0 - throughput: 484.7309743092584 + inference_time: 2025.0 + throughput: 493.82716049382714 estimated_peak_memory_range: - min: 24576 - max: 2098440 + min: 16384 + max: 2363704 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 114 - job_id: j1glwwkjp + job_id: jqp42x78g job_status: Passed torchscript_onnx_qnn: - inference_time: 2008.0 - throughput: 498.00796812749 + inference_time: 2004.0 + throughput: 499.001996007984 estimated_peak_memory_range: - min: 647168 - max: 2343992 + min: 659456 + max: 2021168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 188 - job_id: jygzzzxog + job_id: j2p0x1w9p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:02:32Z' + timestamp: '2024-08-27T00:18:20Z' - torchscript_onnx_tflite: - inference_time: 2018.0 - throughput: 495.5401387512388 + inference_time: 2013.0 + throughput: 496.7709885742673 estimated_peak_memory_range: - min: 40960 - max: 2041120 + min: 32768 + max: 1876200 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 114 - job_id: jw56oo165 + job_id: j0pxz7q35 job_status: Passed torchscript_onnx_qnn: - inference_time: 2032.0 - throughput: 492.12598425196853 + inference_time: 2010.0 + throughput: 497.5124378109453 estimated_peak_memory_range: min: 634880 - max: 1984384 + max: 1817512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 188 - job_id: jz5wyyd3g + job_id: j1p8k3nkp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:02:34Z' + timestamp: '2024-08-27T00:18:21Z' - torchscript_onnx_qnn: - inference_time: 2225.0 - throughput: 449.438202247191 + inference_time: 2182.0 + throughput: 458.29514207149407 estimated_peak_memory_range: - min: 1224704 - max: 1224704 + min: 602112 + max: 602112 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 188 - job_id: j1pv221kg + job_id: jopr74r0g job_status: Passed torchscript_onnx: - inference_time: 2206.0 - throughput: 453.30915684496824 + inference_time: 2211.0 + throughput: 452.2840343735866 estimated_peak_memory_range: - min: 42565632 - max: 42565632 + min: 43118592 + max: 43118592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 190 - job_id: jz57oojvg + job_id: jw5603x65 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:02:38Z' + timestamp: '2024-08-27T00:18:25Z' diff --git a/qai_hub_models/models/regnet_quantized/perf.yaml b/qai_hub_models/models/regnet_quantized/perf.yaml index 2d820365..2f24ce8e 100644 --- a/qai_hub_models/models/regnet_quantized/perf.yaml +++ b/qai_hub_models/models/regnet_quantized/perf.yaml @@ -48,11 +48,11 @@ models: - name: RegNetQuantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 883.0 - throughput: 1132.5028312570782 + inference_time: 871.0 + throughput: 1148.105625717566 estimated_peak_memory_range: - min: 16384 - max: 1455008 + min: 24576 + max: 4037904 primary_compute_unit: NPU precision: int8 layer_info: @@ -60,14 +60,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 116 - job_id: jep2oolrg + job_id: jvgdmryeg job_status: Passed torchscript_onnx_qnn: - inference_time: 1014.0 - throughput: 986.1932938856016 + inference_time: 1009.0 + throughput: 991.0802775024777 estimated_peak_memory_range: - min: 24576 - max: 242618720 + min: 16384 + max: 52610816 primary_compute_unit: NPU precision: int8 layer_info: @@ -75,14 +75,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jwgodd3q5 + job_id: j2p0x186p job_status: Passed torchscript_onnx: - inference_time: 1118.0 - throughput: 894.4543828264758 + inference_time: 1093.0 + throughput: 914.9130832570905 estimated_peak_memory_range: min: 12288 - max: 27274480 + max: 290316912 primary_compute_unit: NPU precision: int8 layer_info: @@ -90,7 +90,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: jz57oodvg + job_id: j7gj806x5 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -99,13 +99,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:01:54Z' + timestamp: '2024-08-27T00:17:46Z' - torchscript_onnx_tflite: - inference_time: 623.0 - throughput: 1605.1364365971108 + inference_time: 626.0 + throughput: 1597.444089456869 estimated_peak_memory_range: min: 12288 - max: 132833248 + max: 135186528 primary_compute_unit: NPU precision: int8 layer_info: @@ -113,14 +113,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 116 - job_id: jqpy8868g + job_id: jz578j1lp job_status: Passed torchscript_onnx_qnn: - inference_time: 732.0 - throughput: 1366.120218579235 + inference_time: 747.0 + throughput: 1338.6880856760374 estimated_peak_memory_range: - min: 163840 - max: 28669792 + min: 159744 + max: 30268944 primary_compute_unit: NPU precision: int8 layer_info: @@ -128,14 +128,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: j1pv22wkg + job_id: j1p8k3dxp job_status: Passed torchscript_onnx: - inference_time: 787.0 - throughput: 1270.6480304955528 + inference_time: 821.0 + throughput: 1218.026796589525 estimated_peak_memory_range: min: 0 - max: 156300704 + max: 158903648 primary_compute_unit: NPU precision: int8 layer_info: @@ -143,7 +143,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: jqp4eew8g + job_id: jlpenr01p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -152,13 +152,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:01:55Z' + timestamp: '2024-08-27T00:17:47Z' - torchscript_onnx_tflite: inference_time: 880.0 throughput: 1136.3636363636363 estimated_peak_memory_range: min: 12288 - max: 2271176 + max: 2181648 primary_compute_unit: NPU precision: int8 layer_info: @@ -166,14 +166,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 116 - job_id: j2p0ool9p + job_id: jqp42x6vg job_status: Passed torchscript_onnx_qnn: - inference_time: 934.0 - throughput: 1070.6638115631692 + inference_time: 938.0 + throughput: 1066.0980810234541 estimated_peak_memory_range: - min: 176128 - max: 1505320 + min: 196608 + max: 1384608 primary_compute_unit: NPU precision: int8 layer_info: @@ -181,7 +181,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jlpe66vog + job_id: jn5qd7x4g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -190,13 +190,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:01:47Z' + timestamp: '2024-08-27T00:17:40Z' - torchscript_onnx_tflite: - inference_time: 1063.0 - throughput: 940.7337723424271 + inference_time: 1038.0 + throughput: 963.3911368015414 estimated_peak_memory_range: - min: 36864 - max: 136858464 + min: 0 + max: 138748736 primary_compute_unit: NPU precision: int8 layer_info: @@ -204,14 +204,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 116 - job_id: j1p8jjzk5 + job_id: j0pxz7815 job_status: Passed torchscript_onnx_qnn: - inference_time: 1189.0 - throughput: 841.0428931875525 + inference_time: 1218.0 + throughput: 821.0180623973728 estimated_peak_memory_range: min: 163840 - max: 31382464 + max: 30564672 primary_compute_unit: NPU precision: int8 layer_info: @@ -219,7 +219,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jnp1oo885 + job_id: jwgo917xg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -228,13 +228,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:01:52Z' + timestamp: '2024-08-27T00:17:44Z' - torchscript_onnx_tflite: - inference_time: 879.0 - throughput: 1137.6564277588168 + inference_time: 875.0 + throughput: 1142.857142857143 estimated_peak_memory_range: - min: 53248 - max: 1586560 + min: 16384 + max: 1577288 primary_compute_unit: NPU precision: int8 layer_info: @@ -242,14 +242,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 116 - job_id: jogk663w5 + job_id: jo5mlw1wg job_status: Passed torchscript_onnx_qnn: - inference_time: 931.0 - throughput: 1074.1138560687432 + inference_time: 940.0 + throughput: 1063.8297872340424 estimated_peak_memory_range: min: 184320 - max: 1411592 + max: 1765752 primary_compute_unit: NPU precision: int8 layer_info: @@ -257,7 +257,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jygzzz7og + job_id: j1glq098p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -266,13 +266,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:01:48Z' + timestamp: '2024-08-27T00:17:41Z' - torchscript_onnx_tflite: - inference_time: 880.0 - throughput: 1136.3636363636363 + inference_time: 875.0 + throughput: 1142.857142857143 estimated_peak_memory_range: - min: 12288 - max: 1344328 + min: 24576 + max: 237156960 primary_compute_unit: NPU precision: int8 layer_info: @@ -280,14 +280,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 116 - job_id: jn5q443ng + job_id: jegnw9drg job_status: Passed torchscript_onnx_qnn: - inference_time: 931.0 - throughput: 1074.1138560687432 + inference_time: 943.0 + throughput: 1060.4453870625662 estimated_peak_memory_range: - min: 184320 - max: 1567648 + min: 176128 + max: 1509008 primary_compute_unit: NPU precision: int8 layer_info: @@ -295,7 +295,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jz5wyy93g + job_id: jw5603905 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -304,13 +304,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:01:49Z' + timestamp: '2024-08-27T00:17:42Z' - torchscript_onnx_tflite: - inference_time: 882.0 - throughput: 1133.7868480725624 + inference_time: 869.0 + throughput: 1150.7479861910242 estimated_peak_memory_range: min: 12288 - max: 3388592 + max: 36765592 primary_compute_unit: NPU precision: int8 layer_info: @@ -318,14 +318,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 116 - job_id: j1glww3jp + job_id: jopr74m9g job_status: Passed torchscript_onnx_qnn: - inference_time: 923.0 - throughput: 1083.4236186348862 + inference_time: 935.0 + throughput: 1069.51871657754 estimated_peak_memory_range: - min: 180224 - max: 1422048 + min: 176128 + max: 1719664 primary_compute_unit: NPU precision: int8 layer_info: @@ -333,7 +333,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jmg9oo4wg + job_id: j1p3r4llp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -342,13 +342,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:01:51Z' + timestamp: '2024-08-27T00:17:43Z' - torchscript_onnx_tflite: - inference_time: 29460.0 - throughput: 33.944331296673454 + inference_time: 29417.0 + throughput: 33.99394907706428 estimated_peak_memory_range: - min: 73728 - max: 72012832 + min: 131072 + max: 76925984 primary_compute_unit: GPU precision: int8 layer_info: @@ -356,14 +356,14 @@ models: layers_on_gpu: 116 layers_on_cpu: 0 total_layers: 116 - job_id: jw56oon65 + job_id: jep2z7q4p job_status: Passed torchscript_onnx_qnn: - inference_time: 3970.0 - throughput: 251.88916876574308 + inference_time: 4053.0 + throughput: 246.7308166790032 estimated_peak_memory_range: - min: 12288 - max: 8254672 + min: 172032 + max: 8295216 primary_compute_unit: NPU precision: int8 layer_info: @@ -371,7 +371,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: jvgd66vrp + job_id: j1pvn1yjg job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -380,13 +380,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:01:53Z' + timestamp: '2024-08-27T00:17:45Z' - torchscript_onnx_tflite: - inference_time: 40904.0 - throughput: 24.44748679835713 + inference_time: 37309.0 + throughput: 26.803184218285132 estimated_peak_memory_range: - min: 765952 - max: 66965552 + min: 565248 + max: 67477512 primary_compute_unit: GPU precision: int8 layer_info: @@ -394,7 +394,7 @@ models: layers_on_gpu: 91 layers_on_cpu: 13 total_layers: 116 - job_id: j1p3ooe3p + job_id: jqpyy4k7p job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -403,13 +403,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T23:01:42Z' + timestamp: '2024-08-27T00:17:36Z' - torchscript_onnx_qnn: - inference_time: 1133.0 - throughput: 882.61253309797 + inference_time: 1220.0 + throughput: 819.672131147541 estimated_peak_memory_range: - min: 475136 - max: 475136 + min: 516096 + max: 516096 primary_compute_unit: NPU precision: int8 layer_info: @@ -417,14 +417,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 113 - job_id: j7gj33lvp + job_id: jogkklw2g job_status: Passed torchscript_onnx: - inference_time: 1122.0 - throughput: 891.2655971479501 + inference_time: 1101.0 + throughput: 908.2652134423251 estimated_peak_memory_range: - min: 24403968 - max: 24403968 + min: 24145920 + max: 24145920 primary_compute_unit: NPU precision: int8 layer_info: @@ -432,7 +432,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 118 - job_id: j0px0013p + job_id: jygz0xqk5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -441,4 +441,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:01:56Z' + timestamp: '2024-08-27T00:17:48Z' diff --git a/qai_hub_models/models/resnet101/perf.yaml b/qai_hub_models/models/resnet101/perf.yaml index 44e2467d..70ce3eea 100644 --- a/qai_hub_models/models/resnet101/perf.yaml +++ b/qai_hub_models/models/resnet101/perf.yaml @@ -45,11 +45,11 @@ models: - name: ResNet101 performance_metrics: - torchscript_onnx_tflite: - inference_time: 3421.0 - throughput: 292.3121894182987 + inference_time: 3440.0 + throughput: 290.69767441860466 estimated_peak_memory_range: - min: 20480 - max: 1738048 + min: 40960 + max: 4293464 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: j2p0oo29p + job_id: jqp42x2vg job_status: Passed torchscript_onnx_qnn: - inference_time: 3468.0 - throughput: 288.35063437139564 + inference_time: 3499.0 + throughput: 285.7959416976279 estimated_peak_memory_range: - min: 626688 - max: 162410136 + min: 622592 + max: 141513816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jwgoddnq5 + job_id: j2p0x1x6p job_status: Passed torchscript_onnx: - inference_time: 3545.0 - throughput: 282.08744710860367 + inference_time: 3600.0 + throughput: 277.77777777777777 estimated_peak_memory_range: min: 16384 - max: 102542136 + max: 102157712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 247 - job_id: jvgd66orp + job_id: j1pvn1njg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:00:46Z' + timestamp: '2024-08-27T00:16:44Z' - torchscript_onnx_tflite: - inference_time: 2435.0 - throughput: 410.6776180698152 + inference_time: 2439.0 + throughput: 410.0041000410004 estimated_peak_memory_range: min: 16384 - max: 114106768 + max: 114847232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: j1p8jjmk5 + job_id: j0pxz7z15 job_status: Passed torchscript_onnx_qnn: - inference_time: 2512.0 - throughput: 398.0891719745223 + inference_time: 2525.0 + throughput: 396.03960396039605 estimated_peak_memory_range: - min: 0 - max: 34073696 + min: 618496 + max: 37854256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: j1pv22vkg + job_id: j1p8k3kxp job_status: Passed torchscript_onnx: - inference_time: 2583.0 - throughput: 387.14672861014327 + inference_time: 2600.0 + throughput: 384.61538461538464 estimated_peak_memory_range: - min: 606208 - max: 117660272 + min: 0 + max: 118305152 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 247 - job_id: jz57ooxvg + job_id: j7gj808x5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:00:47Z' + timestamp: '2024-08-27T00:16:45Z' - torchscript_onnx_tflite: - inference_time: 3365.0 - throughput: 297.1768202080238 + inference_time: 3383.0 + throughput: 295.5956251847473 estimated_peak_memory_range: min: 16384 - max: 44786248 + max: 2309104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jogk66qw5 + job_id: jo5mlwlwg job_status: Passed torchscript_onnx_qnn: - inference_time: 3296.0 - throughput: 303.3980582524272 + inference_time: 3293.0 + throughput: 303.67446097783176 estimated_peak_memory_range: - min: 638976 - max: 1986720 + min: 647168 + max: 2345760 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jlpe66kog + job_id: jn5qd7d4g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T23:00:40Z' + timestamp: '2024-08-27T00:16:39Z' - torchscript_onnx_tflite: - inference_time: 4808.0 - throughput: 207.98668885191347 + inference_time: 4769.0 + throughput: 209.68756552736423 estimated_peak_memory_range: - min: 20480 - max: 94878272 + min: 16384 + max: 94350128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jn5q44rng + job_id: jegnw9wrg job_status: Passed torchscript_onnx_qnn: - inference_time: 4901.0 - throughput: 204.03999183840034 + inference_time: 4850.0 + throughput: 206.18556701030928 estimated_peak_memory_range: - min: 622592 - max: 22235008 + min: 618496 + max: 24073424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jnp1ooe85 + job_id: jwgo919xg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:00:45Z' + timestamp: '2024-08-27T00:16:43Z' - torchscript_onnx_tflite: - inference_time: 3420.0 - throughput: 292.39766081871346 + inference_time: 3411.0 + throughput: 293.1691586045148 estimated_peak_memory_range: - min: 32768 - max: 2212968 + min: 20480 + max: 103602832 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: j1glww2jp + job_id: jopr7479g job_status: Passed torchscript_onnx_qnn: - inference_time: 3292.0 - throughput: 303.7667071688943 + inference_time: 3283.0 + throughput: 304.5994517209869 estimated_peak_memory_range: min: 634880 - max: 2196416 + max: 1894304 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jygzzzrog + job_id: j1glq0q8p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T23:00:41Z' + timestamp: '2024-08-27T00:16:40Z' - torchscript_onnx_tflite: - inference_time: 3375.0 - throughput: 296.2962962962963 + inference_time: 3357.0 + throughput: 297.8850163836759 estimated_peak_memory_range: min: 24576 - max: 1767360 + max: 2039880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jw56ooz65 + job_id: jep2z7z4p job_status: Passed torchscript_onnx_qnn: - inference_time: 3297.0 - throughput: 303.3060357901122 + inference_time: 3319.0 + throughput: 301.29557095510694 estimated_peak_memory_range: - min: 630784 - max: 1995008 + min: 679936 + max: 2117808 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jz5wyyq3g + job_id: jw5603005 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T23:00:42Z' + timestamp: '2024-08-27T00:16:41Z' - torchscript_onnx_tflite: - inference_time: 3370.0 - throughput: 296.7359050445104 + inference_time: 3366.0 + throughput: 297.08853238265004 estimated_peak_memory_range: - min: 16384 - max: 1962184 + min: 28672 + max: 2380736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: j1p3oo13p + job_id: jqpyy4y7p job_status: Passed torchscript_onnx_qnn: - inference_time: 3347.0 - throughput: 298.7750224081267 + inference_time: 3343.0 + throughput: 299.1325157044571 estimated_peak_memory_range: - min: 643072 - max: 1866144 + min: 647168 + max: 2230880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jmg9oowwg + job_id: j1p3r4rlp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T23:00:43Z' + timestamp: '2024-08-27T00:16:42Z' - torchscript_onnx_qnn: - inference_time: 3460.0 - throughput: 289.01734104046244 + inference_time: 3492.0 + throughput: 286.368843069874 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: j7gj33evp + job_id: jogkklk2g job_status: Passed torchscript_onnx: - inference_time: 3501.0 - throughput: 285.6326763781777 + inference_time: 3718.0 + throughput: 268.9618074233459 estimated_peak_memory_range: - min: 92049408 - max: 92049408 + min: 91951104 + max: 91951104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 247 - job_id: jqp4eev8g + job_id: jlpenrn1p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:00:48Z' + timestamp: '2024-08-27T00:16:46Z' diff --git a/qai_hub_models/models/resnet101_quantized/perf.yaml b/qai_hub_models/models/resnet101_quantized/perf.yaml index ca631864..210c5ff7 100644 --- a/qai_hub_models/models/resnet101_quantized/perf.yaml +++ b/qai_hub_models/models/resnet101_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: ResNet101Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 1170.0 - throughput: 854.7008547008547 + inference_time: 1149.0 + throughput: 870.3220191470845 estimated_peak_memory_range: min: 12288 - max: 6467648 + max: 19863256 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jn5q44e4g + job_id: jz578jnlp job_status: Passed torchscript_onnx_qnn: inference_time: 1377.0 throughput: 726.2164124909223 estimated_peak_memory_range: - min: 16384 - max: 36154152 + min: 61440 + max: 58914904 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jz5wyy36g + job_id: j1p8k3rxp job_status: Passed torchscript_onnx: - inference_time: 1459.0 - throughput: 685.4009595613434 + inference_time: 1468.0 + throughput: 681.1989100817439 estimated_peak_memory_range: min: 16384 - max: 51997216 + max: 194040200 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 151 - job_id: jqp4eed8g + job_id: jlpenrx1p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T23:00:03Z' + timestamp: '2024-08-27T00:16:04Z' - torchscript_onnx_tflite: - inference_time: 901.0 - throughput: 1109.8779134295228 + inference_time: 931.0 + throughput: 1074.1138560687432 estimated_peak_memory_range: - min: 12288 - max: 99127040 + min: 16384 + max: 101233440 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: j1glww68p + job_id: jqp42x4vg job_status: Passed torchscript_onnx_qnn: - inference_time: 1044.0 - throughput: 957.8544061302682 + inference_time: 1046.0 + throughput: 956.0229445506692 estimated_peak_memory_range: - min: 163840 - max: 21752192 + min: 167936 + max: 24405840 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jmg9ooylg + job_id: jogkkl02g job_status: Passed torchscript_onnx: - inference_time: 1114.0 - throughput: 897.6660682226212 + inference_time: 1135.0 + throughput: 881.0572687224669 estimated_peak_memory_range: min: 0 - max: 117455792 + max: 118765056 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 151 - job_id: j0px0063p + job_id: jygz0xyk5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T23:00:04Z' + timestamp: '2024-08-27T00:16:04Z' - torchscript_onnx_tflite: - inference_time: 1166.0 - throughput: 857.6329331046312 + inference_time: 1150.0 + throughput: 869.5652173913044 estimated_peak_memory_range: min: 12288 - max: 1500792 + max: 5520176 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jw56ooe05 + job_id: j0pxz7r15 job_status: Passed torchscript_onnx_qnn: - inference_time: 1320.0 - throughput: 757.5757575757576 + inference_time: 1321.0 + throughput: 757.002271006813 estimated_peak_memory_range: - min: 180224 - max: 1349496 + min: 184320 + max: 1709856 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jvgd66qep + job_id: j1glq088p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:59:55Z' + timestamp: '2024-08-27T00:15:58Z' - torchscript_onnx_tflite: - inference_time: 1388.0 - throughput: 720.4610951008646 + inference_time: 1360.0 + throughput: 735.2941176470588 estimated_peak_memory_range: - min: 12288 - max: 99438704 + min: 16384 + max: 100420688 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: j1p3oovlp + job_id: jo5mlwkwg job_status: Passed torchscript_onnx_qnn: - inference_time: 1587.0 - throughput: 630.119722747322 + inference_time: 1596.0 + throughput: 626.5664160401003 estimated_peak_memory_range: - min: 163840 - max: 22982576 + min: 167936 + max: 24190688 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jvgd66qrp + job_id: j1pvn1mjg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T23:00:00Z' + timestamp: '2024-08-27T00:16:02Z' - torchscript_onnx_tflite: - inference_time: 1162.0 - throughput: 860.5851979345955 + inference_time: 1153.0 + throughput: 867.3026886383348 estimated_peak_memory_range: - min: 16384 - max: 23185936 + min: 28672 + max: 352430736 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jwgoddkx5 + job_id: jegnw9qrg job_status: Passed torchscript_onnx_qnn: - inference_time: 1323.0 - throughput: 755.8578987150415 + inference_time: 1329.0 + throughput: 752.4454477050414 estimated_peak_memory_range: - min: 184320 - max: 1512256 + min: 172032 + max: 1682152 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jz5wyy33g + job_id: jw5603m05 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:59:56Z' + timestamp: '2024-08-27T00:15:59Z' - torchscript_onnx_tflite: - inference_time: 1184.0 - throughput: 844.5945945945946 + inference_time: 1155.0 + throughput: 865.8008658008658 estimated_peak_memory_range: - min: 36864 - max: 33261152 + min: 40960 + max: 14167216 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: j1pv220jg + job_id: jopr74d9g job_status: Passed torchscript_onnx_qnn: - inference_time: 1327.0 - throughput: 753.5795026375282 + inference_time: 1332.0 + throughput: 750.7507507507507 estimated_peak_memory_range: - min: 176128 - max: 1395760 + min: 180224 + max: 1475888 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jmg9ooywg + job_id: j1p3r47lp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:59:58Z' + timestamp: '2024-08-27T00:16:00Z' - torchscript_onnx_tflite: - inference_time: 1178.0 - throughput: 848.8964346349745 + inference_time: 1156.0 + throughput: 865.0519031141869 estimated_peak_memory_range: - min: 28672 - max: 13264920 + min: 49152 + max: 393922808 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: j7gj33zxp + job_id: jep2z7d4p job_status: Passed torchscript_onnx_qnn: - inference_time: 1329.0 - throughput: 752.4454477050414 + inference_time: 1325.0 + throughput: 754.7169811320755 estimated_peak_memory_range: - min: 172032 - max: 1488152 + min: 176128 + max: 1401776 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jnp1oow85 + job_id: jwgo91wxg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:59:59Z' + timestamp: '2024-08-27T00:16:01Z' - torchscript_onnx_tflite: - inference_time: 4598.0 - throughput: 217.48586341887778 + inference_time: 4491.0 + throughput: 222.667557336896 estimated_peak_memory_range: - min: 20480 - max: 33214304 + min: 12288 + max: 34986864 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jlpe66e1g + job_id: jqpyy427p job_status: Passed torchscript_onnx_qnn: - inference_time: 6384.0 - throughput: 156.64160401002508 + inference_time: 6294.0 + throughput: 158.8814744200826 estimated_peak_memory_range: min: 176128 - max: 8071344 + max: 8620608 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jz57oolvg + job_id: j7gj80yx5 job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T23:00:02Z' + timestamp: '2024-08-27T00:16:03Z' - torchscript_onnx_tflite: - inference_time: 17269.0 - throughput: 57.90723261335341 + inference_time: 17193.0 + throughput: 58.16320595591229 estimated_peak_memory_range: - min: 40960 - max: 2157072 + min: 36864 + max: 9801104 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jygzzzokg + job_id: j2p0x196p job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:59:50Z' + timestamp: '2024-08-27T00:15:53Z' - torchscript_onnx_qnn: - inference_time: 1321.0 - throughput: 757.002271006813 + inference_time: 1365.0 + throughput: 732.6007326007326 estimated_peak_memory_range: - min: 360448 - max: 360448 + min: 1585152 + max: 1585152 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jnp1oow25 + job_id: jn5qd714g job_status: Passed torchscript_onnx: - inference_time: 1332.0 - throughput: 750.7507507507507 + inference_time: 1506.0 + throughput: 664.0106241699867 estimated_peak_memory_range: - min: 50720768 - max: 50720768 + min: 50642944 + max: 50642944 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 151 - job_id: jo5m996dg + job_id: jz5wrdr6p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T23:00:05Z' + timestamp: '2024-08-27T00:16:05Z' diff --git a/qai_hub_models/models/resnet18/perf.yaml b/qai_hub_models/models/resnet18/perf.yaml index de75f6b2..5a983286 100644 --- a/qai_hub_models/models/resnet18/perf.yaml +++ b/qai_hub_models/models/resnet18/perf.yaml @@ -45,11 +45,11 @@ models: - name: ResNet18 performance_metrics: - torchscript_onnx_tflite: - inference_time: 1415.0 - throughput: 706.7137809187279 + inference_time: 1386.0 + throughput: 721.5007215007215 estimated_peak_memory_range: min: 12288 - max: 357282536 + max: 1947232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 38 - job_id: jw56oo205 + job_id: jvgdmr6eg job_status: Passed torchscript_onnx_qnn: - inference_time: 1473.0 - throughput: 678.8866259334691 + inference_time: 1449.0 + throughput: 690.1311249137336 estimated_peak_memory_range: - min: 12288 - max: 93946944 + min: 16384 + max: 62603632 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jz5wyy26g + job_id: jep2z7o4p job_status: Passed torchscript_onnx: - inference_time: 1337.0 - throughput: 747.9431563201197 + inference_time: 1332.0 + throughput: 750.7507507507507 estimated_peak_memory_range: - min: 36864 - max: 369031024 + min: 16384 + max: 25871720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 55 - job_id: jegn11lrp + job_id: jwgo91dxg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:58:04Z' + timestamp: '2024-08-27T00:14:10Z' - torchscript_onnx_tflite: - inference_time: 972.0 - throughput: 1028.80658436214 + inference_time: 971.0 + throughput: 1029.8661174047375 estimated_peak_memory_range: min: 16384 - max: 27014640 + max: 27804288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 38 - job_id: j1p3oonlp + job_id: jz578jolp job_status: Passed torchscript_onnx_qnn: - inference_time: 1011.0 - throughput: 989.1196834817013 + inference_time: 1010.0 + throughput: 990.0990099009902 estimated_peak_memory_range: - min: 0 - max: 15536160 + min: 618496 + max: 17096400 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jmg9oojlg + job_id: jqpyy487p job_status: Passed torchscript_onnx: - inference_time: 991.0 - throughput: 1009.0817356205853 + inference_time: 992.0 + throughput: 1008.0645161290323 estimated_peak_memory_range: min: 0 - max: 28025680 + max: 28273584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 55 - job_id: joprxx89p + job_id: j1pvn12jg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:58:05Z' + timestamp: '2024-08-27T00:14:11Z' - torchscript_onnx_tflite: - inference_time: 1422.0 - throughput: 703.2348804500704 + inference_time: 1381.0 + throughput: 724.112961622013 estimated_peak_memory_range: - min: 28672 - max: 1801080 + min: 12288 + max: 1305128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 38 - job_id: jwgoddzx5 + job_id: jqp42xevg job_status: Passed torchscript_onnx_qnn: - inference_time: 1460.0 - throughput: 684.931506849315 + inference_time: 1452.0 + throughput: 688.7052341597796 estimated_peak_memory_range: - min: 634880 - max: 2073640 + min: 630784 + max: 2204880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jvgd66eep + job_id: j1p8k3jxp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:57:58Z' + timestamp: '2024-08-27T00:14:05Z' - torchscript_onnx_tflite: - inference_time: 1978.0 - throughput: 505.5611729019211 + inference_time: 1934.0 + throughput: 517.063081695967 estimated_peak_memory_range: - min: 16384 - max: 27551424 + min: 24576 + max: 27697936 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 38 - job_id: j1pv22qjg + job_id: j0pxz7015 job_status: Passed torchscript_onnx_qnn: - inference_time: 1994.0 - throughput: 501.5045135406219 + inference_time: 1989.0 + throughput: 502.76520864756156 estimated_peak_memory_range: min: 618496 - max: 16975376 + max: 19486480 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jo5m99qwg + job_id: j1p3r4olp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:58:03Z' + timestamp: '2024-08-27T00:14:09Z' - torchscript_onnx_tflite: - inference_time: 1416.0 - throughput: 706.2146892655368 + inference_time: 1385.0 + throughput: 722.0216606498195 estimated_peak_memory_range: - min: 20480 - max: 11689840 + min: 24576 + max: 1780720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 38 - job_id: j7gj33dxp + job_id: jo5mlw9wg job_status: Passed torchscript_onnx_qnn: - inference_time: 1467.0 - throughput: 681.6632583503749 + inference_time: 1461.0 + throughput: 684.4626967830253 estimated_peak_memory_range: - min: 651264 - max: 1968400 + min: 634880 + max: 2162232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jz57oo0lg + job_id: jogkkl62g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:57:59Z' + timestamp: '2024-08-27T00:14:06Z' - torchscript_onnx_tflite: - inference_time: 1408.0 - throughput: 710.2272727272727 + inference_time: 1393.0 + throughput: 717.8750897343862 estimated_peak_memory_range: - min: 28672 - max: 34018848 + min: 61440 + max: 2017280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 38 - job_id: jlpe66o1g + job_id: jegnw91rg job_status: Passed torchscript_onnx_qnn: - inference_time: 1466.0 - throughput: 682.1282401091405 + inference_time: 1456.0 + throughput: 686.8131868131868 estimated_peak_memory_range: - min: 655360 - max: 1850968 + min: 634880 + max: 2086048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jqp4eekvg + job_id: j1glq0w8p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:58:00Z' + timestamp: '2024-08-27T00:14:07Z' - torchscript_onnx_tflite: - inference_time: 1419.0 - throughput: 704.7216349541931 + inference_time: 1383.0 + throughput: 723.0657989877079 estimated_peak_memory_range: - min: 32768 - max: 1491384 + min: 28672 + max: 290541624 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 38 - job_id: jygzzz2kg + job_id: jopr74x9g job_status: Passed torchscript_onnx_qnn: - inference_time: 1478.0 - throughput: 676.5899864682003 + inference_time: 1456.0 + throughput: 686.8131868131868 estimated_peak_memory_range: - min: 634880 - max: 2105632 + min: 643072 + max: 1831000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: j0px00n1p + job_id: jw5603o05 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:58:02Z' + timestamp: '2024-08-27T00:14:08Z' - torchscript_onnx_qnn: - inference_time: 1565.0 - throughput: 638.9776357827476 + inference_time: 1572.0 + throughput: 636.1323155216285 estimated_peak_memory_range: - min: 606208 - max: 606208 + min: 602112 + max: 602112 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jnp1ooy25 + job_id: j2p0x1o6p job_status: Passed torchscript_onnx: - inference_time: 1308.0 - throughput: 764.525993883792 + inference_time: 1341.0 + throughput: 745.7121551081283 estimated_peak_memory_range: - min: 24952832 - max: 24952832 + min: 25440256 + max: 25440256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 55 - job_id: jep2oo04g + job_id: j7gj803x5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:58:06Z' + timestamp: '2024-08-27T00:14:12Z' diff --git a/qai_hub_models/models/resnet18_quantized/perf.yaml b/qai_hub_models/models/resnet18_quantized/perf.yaml index 44b0b355..c6bc9dba 100644 --- a/qai_hub_models/models/resnet18_quantized/perf.yaml +++ b/qai_hub_models/models/resnet18_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: ResNet18Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 433.0 - throughput: 2309.4688221709007 + inference_time: 404.0 + throughput: 2475.2475247524753 estimated_peak_memory_range: - min: 53248 - max: 38491776 + min: 12288 + max: 121975024 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j1glwwv8p + job_id: jz578jerp job_status: Passed torchscript_onnx_qnn: - inference_time: 625.0 - throughput: 1600.0 + inference_time: 640.0 + throughput: 1562.5 estimated_peak_memory_range: - min: 0 - max: 126764368 + min: 16384 + max: 126986680 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: jmg9oo0lg + job_id: j1p8k3e8p job_status: Passed torchscript_onnx: - inference_time: 639.0 - throughput: 1564.9452269170579 + inference_time: 642.0 + throughput: 1557.632398753894 estimated_peak_memory_range: - min: 12288 - max: 23127664 + min: 36864 + max: 14043416 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jep2oox4g + job_id: jygz0xl65 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:57:25Z' + timestamp: '2024-08-27T00:13:34Z' - torchscript_onnx_tflite: - inference_time: 322.0 - throughput: 3105.590062111801 + inference_time: 317.0 + throughput: 3154.5741324921137 estimated_peak_memory_range: min: 12288 - max: 27076416 + max: 27848608 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jw56ooy05 + job_id: jqp42xylg job_status: Passed torchscript_onnx_qnn: - inference_time: 480.0 - throughput: 2083.3333333333335 + inference_time: 479.0 + throughput: 2087.6826722338205 estimated_peak_memory_range: - min: 163840 - max: 11781824 + min: 159744 + max: 13688240 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: jnp1oo225 + job_id: jogkkl2og job_status: Passed torchscript_onnx: - inference_time: 515.0 - throughput: 1941.7475728155339 + inference_time: 529.0 + throughput: 1890.359168241966 estimated_peak_memory_range: min: 12288 - max: 30556064 + max: 31589520 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: jqpy88z7g + job_id: jz5wrdljp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:57:26Z' + timestamp: '2024-08-27T00:13:35Z' - torchscript_onnx_tflite: - inference_time: 422.0 - throughput: 2369.6682464454975 + inference_time: 405.0 + throughput: 2469.135802469136 estimated_peak_memory_range: - min: 12288 - max: 1354496 + min: 32768 + max: 1453368 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j1p3oojlp + job_id: j0pxz7l95 job_status: Passed torchscript_onnx_qnn: - inference_time: 632.0 - throughput: 1582.2784810126582 + inference_time: 628.0 + throughput: 1592.3566878980891 estimated_peak_memory_range: - min: 180224 - max: 1470984 + min: 184320 + max: 1788320 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: jz57oo2lg + job_id: j1glq0ylp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:57:18Z' + timestamp: '2024-08-27T00:13:29Z' - torchscript_onnx_tflite: - inference_time: 508.0 - throughput: 1968.5039370078741 + inference_time: 472.0 + throughput: 2118.64406779661 estimated_peak_memory_range: min: 16384 - max: 27790192 + max: 28580096 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jwgodd2x5 + job_id: jo5mlw0qg job_status: Passed torchscript_onnx_qnn: - inference_time: 709.0 - throughput: 1410.4372355430182 + inference_time: 700.0 + throughput: 1428.5714285714287 estimated_peak_memory_range: - min: 0 - max: 14574928 + min: 159744 + max: 13858896 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: jegn110rp + job_id: j7gj80r85 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:57:23Z' + timestamp: '2024-08-27T00:13:32Z' - torchscript_onnx_tflite: - inference_time: 416.0 - throughput: 2403.846153846154 + inference_time: 404.0 + throughput: 2475.2475247524753 estimated_peak_memory_range: - min: 28672 - max: 1378408 + min: 12288 + max: 7032328 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j1pv226jg + job_id: jegnw9zmg job_status: Passed torchscript_onnx_qnn: - inference_time: 636.0 - throughput: 1572.3270440251572 + inference_time: 629.0 + throughput: 1589.825119236884 estimated_peak_memory_range: - min: 180224 - max: 1690016 + min: 184320 + max: 1622288 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: jqp4eenvg + job_id: jw5603875 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:57:19Z' + timestamp: '2024-08-27T00:13:30Z' - torchscript_onnx_tflite: - inference_time: 407.0 - throughput: 2457.002457002457 + inference_time: 405.0 + throughput: 2469.135802469136 estimated_peak_memory_range: - min: 0 - max: 11657632 + min: 45056 + max: 14625128 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j7gj33vxp + job_id: jopr74leg job_status: Passed torchscript_onnx_qnn: - inference_time: 632.0 - throughput: 1582.2784810126582 + inference_time: 630.0 + throughput: 1587.3015873015872 estimated_peak_memory_range: - min: 188416 - max: 1455840 + min: 184320 + max: 1450456 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: j0px0091p + job_id: j1p3r4zzp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:57:20Z' + timestamp: '2024-08-27T00:13:31Z' - torchscript_onnx_tflite: - inference_time: 438.0 - throughput: 2283.10502283105 + inference_time: 409.0 + throughput: 2444.987775061125 estimated_peak_memory_range: - min: 49152 - max: 4903344 + min: 16384 + max: 1410744 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jlpe66d1g + job_id: jep2z7rmp job_status: Passed torchscript_onnx_qnn: - inference_time: 631.0 - throughput: 1584.7860538827258 + inference_time: 629.0 + throughput: 1589.825119236884 estimated_peak_memory_range: - min: 184320 - max: 1391784 + min: 188416 + max: 1780360 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: jo5m99ewg + job_id: j1pvn1lmg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:57:22Z' + timestamp: '2024-08-27T00:13:32Z' - torchscript_onnx_tflite: - inference_time: 1413.0 - throughput: 707.7140835102618 + inference_time: 1358.0 + throughput: 736.3770250368188 estimated_peak_memory_range: - min: 16384 - max: 17204896 + min: 12288 + max: 18022176 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jygzzz3kg + job_id: jqpyy4o4p job_status: Passed torchscript_onnx_qnn: - inference_time: 2038.0 - throughput: 490.6771344455348 + inference_time: 2097.0 + throughput: 476.87172150691464 estimated_peak_memory_range: min: 12288 - max: 8266560 + max: 7916624 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: joprxx69p + job_id: jlpenr70p job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:57:24Z' + timestamp: '2024-08-27T00:13:33Z' - torchscript_onnx_tflite: - inference_time: 7068.0 - throughput: 141.4827391058291 + inference_time: 7112.0 + throughput: 140.607424071991 estimated_peak_memory_range: - min: 28672 - max: 6241056 + min: 12288 + max: 1819616 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jz5wyyw6g + job_id: j2p0x1mep job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:57:13Z' + timestamp: '2024-08-27T00:13:25Z' - torchscript_onnx_qnn: - inference_time: 716.0 - throughput: 1396.6480446927374 + inference_time: 700.0 + throughput: 1428.5714285714287 estimated_peak_memory_range: - min: 466944 - max: 466944 + min: 495616 + max: 495616 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 37 - job_id: jvgd66nep + job_id: jn5qd7lmg job_status: Passed torchscript_onnx: - inference_time: 649.0 - throughput: 1540.8320493066255 + inference_time: 864.0 + throughput: 1157.4074074074074 estimated_peak_memory_range: - min: 14360576 - max: 14360576 + min: 15843328 + max: 15843328 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 42 - job_id: j2p0oo46p + job_id: jmg9q3zvp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:57:27Z' + timestamp: '2024-08-27T00:13:37Z' diff --git a/qai_hub_models/models/resnet50/perf.yaml b/qai_hub_models/models/resnet50/perf.yaml index 802a90a4..e84877a8 100644 --- a/qai_hub_models/models/resnet50/perf.yaml +++ b/qai_hub_models/models/resnet50/perf.yaml @@ -45,11 +45,11 @@ models: - name: ResNet50 performance_metrics: - torchscript_onnx_tflite: - inference_time: 2290.0 - throughput: 436.68122270742356 + inference_time: 2259.0 + throughput: 442.67374944665784 estimated_peak_memory_range: min: 16384 - max: 2117776 + max: 2127688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: j1p3oo0lp + job_id: jqp42x1lg job_status: Passed torchscript_onnx_qnn: - inference_time: 2383.0 - throughput: 419.639110365086 + inference_time: 2388.0 + throughput: 418.7604690117253 estimated_peak_memory_range: min: 622592 - max: 185362176 + max: 185010736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jmg9oollg + job_id: j2p0x1jep job_status: Passed torchscript_onnx: - inference_time: 2354.0 - throughput: 424.8088360237893 + inference_time: 2350.0 + throughput: 425.531914893617 estimated_peak_memory_range: - min: 32768 - max: 59924592 + min: 12288 + max: 59664704 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: joprxxv9p + job_id: j7gj80785 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:56:28Z' + timestamp: '2024-08-27T00:12:41Z' - torchscript_onnx_tflite: - inference_time: 1603.0 - throughput: 623.8303181534623 + inference_time: 1611.0 + throughput: 620.7324643078833 estimated_peak_memory_range: min: 16384 - max: 78104752 + max: 77540928 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jwgodd6x5 + job_id: j0pxz7495 job_status: Passed torchscript_onnx_qnn: - inference_time: 1698.0 - throughput: 588.9281507656066 + inference_time: 1684.0 + throughput: 593.8242280285035 estimated_peak_memory_range: - min: 0 - max: 27036496 + min: 618496 + max: 27001056 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jnp1oo425 + job_id: j1p8k3x8p job_status: Passed torchscript_onnx: - inference_time: 1755.0 - throughput: 569.8005698005697 + inference_time: 1729.0 + throughput: 578.368999421631 estimated_peak_memory_range: min: 0 - max: 78028688 + max: 78149600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jep2ook4g + job_id: jlpenrz0p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:56:29Z' + timestamp: '2024-08-27T00:12:43Z' - torchscript_onnx_tflite: - inference_time: 2265.0 - throughput: 441.5011037527594 + inference_time: 2248.0 + throughput: 444.83985765124555 estimated_peak_memory_range: - min: 12288 - max: 42862256 + min: 32768 + max: 64115896 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: j1pv22kjg + job_id: jo5mlwmqg job_status: Passed torchscript_onnx_qnn: - inference_time: 2199.0 - throughput: 454.7521600727603 + inference_time: 2169.0 + throughput: 461.04195481788844 estimated_peak_memory_range: - min: 630784 - max: 1794064 + min: 634880 + max: 1803840 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jz57ooylg + job_id: jn5qd7ymg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:56:22Z' + timestamp: '2024-08-27T00:12:36Z' - torchscript_onnx_tflite: - inference_time: 3125.0 - throughput: 320.0 + inference_time: 3091.0 + throughput: 323.51989647363314 estimated_peak_memory_range: - min: 20480 - max: 64749856 + min: 73728 + max: 66028736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: j7gj33nxp + job_id: jegnw9nmg job_status: Passed torchscript_onnx_qnn: - inference_time: 3137.0 - throughput: 318.77590054191904 + inference_time: 3131.0 + throughput: 319.38677738741615 estimated_peak_memory_range: min: 618496 - max: 21513424 + max: 18789824 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jegn116rp + job_id: j1pvn1dmg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:56:27Z' + timestamp: '2024-08-27T00:12:40Z' - torchscript_onnx_tflite: - inference_time: 2292.0 - throughput: 436.3001745200698 + inference_time: 2247.0 + throughput: 445.0378282153983 estimated_peak_memory_range: - min: 180224 - max: 1743640 + min: 24576 + max: 2461968 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jlpe66m1g + job_id: jopr740eg job_status: Passed torchscript_onnx_qnn: - inference_time: 2205.0 - throughput: 453.51473922902494 + inference_time: 2177.0 + throughput: 459.34772622875516 estimated_peak_memory_range: - min: 626688 - max: 2203960 + min: 643072 + max: 1931656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jqp4eelvg + job_id: jw5603775 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:56:23Z' + timestamp: '2024-08-27T00:12:37Z' - torchscript_onnx_tflite: - inference_time: 2281.0 - throughput: 438.4042086804033 + inference_time: 2255.0 + throughput: 443.4589800443459 estimated_peak_memory_range: - min: 12288 - max: 2076936 + min: 24576 + max: 1666936 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jygzzzdkg + job_id: jep2z7wmp job_status: Passed torchscript_onnx_qnn: - inference_time: 2203.0 - throughput: 453.92646391284615 + inference_time: 2182.0 + throughput: 458.29514207149407 estimated_peak_memory_range: - min: 638976 - max: 1940664 + min: 630784 + max: 1907640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j0px00k1p + job_id: j1p3r49zp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:56:24Z' + timestamp: '2024-08-27T00:12:38Z' - torchscript_onnx_tflite: - inference_time: 2278.0 - throughput: 438.98156277436345 + inference_time: 2261.0 + throughput: 442.2821760283061 estimated_peak_memory_range: - min: 16384 - max: 2051432 + min: 32768 + max: 2045560 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jz5wyye6g + job_id: jqpyy4x4p job_status: Passed torchscript_onnx_qnn: - inference_time: 2256.0 - throughput: 443.26241134751774 + inference_time: 2204.0 + throughput: 453.7205081669691 estimated_peak_memory_range: - min: 634880 - max: 1982000 + min: 630784 + max: 1848272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jo5m99nwg + job_id: jwgo91rdg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:56:26Z' + timestamp: '2024-08-27T00:12:39Z' - torchscript_onnx_qnn: - inference_time: 2307.0 - throughput: 433.4633723450368 + inference_time: 2368.0 + throughput: 422.2972972972973 estimated_peak_memory_range: - min: 606208 - max: 606208 + min: 991232 + max: 991232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jvgd66xep + job_id: jogkkl4og job_status: Passed torchscript_onnx: - inference_time: 2310.0 - throughput: 432.9004329004329 + inference_time: 2384.0 + throughput: 419.46308724832215 estimated_peak_memory_range: - min: 54841344 - max: 54841344 + min: 54681600 + max: 54681600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jqpy8817g + job_id: jygz0xm65 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:56:30Z' + timestamp: '2024-08-27T00:12:44Z' diff --git a/qai_hub_models/models/resnet50_quantized/perf.yaml b/qai_hub_models/models/resnet50_quantized/perf.yaml index acb25fd9..d8ce569c 100644 --- a/qai_hub_models/models/resnet50_quantized/perf.yaml +++ b/qai_hub_models/models/resnet50_quantized/perf.yaml @@ -48,11 +48,11 @@ models: - name: ResNet50Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 813.0 - throughput: 1230.0123001230013 + inference_time: 786.0 + throughput: 1272.264631043257 estimated_peak_memory_range: - min: 20480 - max: 19632528 + min: 12288 + max: 272062552 primary_compute_unit: NPU precision: int8 layer_info: @@ -60,14 +60,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: j7gj3r48p + job_id: jvgdmrzlg job_status: Passed torchscript_onnx_qnn: - inference_time: 1003.0 - throughput: 997.0089730807578 + inference_time: 1006.0 + throughput: 994.0357852882704 estimated_peak_memory_range: - min: 24576 - max: 33049952 + min: 16384 + max: 35159360 primary_compute_unit: NPU precision: int8 layer_info: @@ -75,14 +75,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jnp1ooz25 + job_id: j1p8k318p job_status: Passed torchscript_onnx: - inference_time: 1039.0 - throughput: 962.4639076034649 + inference_time: 1057.0 + throughput: 946.073793755913 estimated_peak_memory_range: min: 16384 - max: 31157896 + max: 314692016 primary_compute_unit: NPU precision: int8 layer_info: @@ -90,7 +90,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: jqpy8837g + job_id: jlpenry0p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -99,13 +99,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:55:47Z' + timestamp: '2024-08-27T00:12:03Z' - torchscript_onnx_tflite: - inference_time: 590.0 - throughput: 1694.915254237288 + inference_time: 605.0 + throughput: 1652.892561983471 estimated_peak_memory_range: min: 12288 - max: 63633984 + max: 63629808 primary_compute_unit: NPU precision: int8 layer_info: @@ -113,14 +113,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jlpe6730g + job_id: jz578j9rp job_status: Passed torchscript_onnx_qnn: - inference_time: 765.0 - throughput: 1307.18954248366 + inference_time: 760.0 + throughput: 1315.7894736842106 estimated_peak_memory_range: min: 167936 - max: 16367712 + max: 17928688 primary_compute_unit: NPU precision: int8 layer_info: @@ -128,14 +128,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jvgd661ep + job_id: jogkkl8og job_status: Passed torchscript_onnx: - inference_time: 838.0 - throughput: 1193.3174224343675 + inference_time: 820.0 + throughput: 1219.5121951219512 estimated_peak_memory_range: - min: 0 - max: 79509152 + min: 28672 + max: 79259936 primary_compute_unit: NPU precision: int8 layer_info: @@ -143,7 +143,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: j2p0oo06p + job_id: jygz0xn65 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -152,13 +152,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:55:48Z' + timestamp: '2024-08-27T00:12:04Z' - torchscript_onnx_tflite: - inference_time: 788.0 - throughput: 1269.0355329949239 + inference_time: 776.0 + throughput: 1288.659793814433 estimated_peak_memory_range: - min: 28672 - max: 17387840 + min: 12288 + max: 12873432 primary_compute_unit: NPU precision: int8 layer_info: @@ -166,14 +166,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jygzzlk6g + job_id: jqp42x3lg job_status: Passed torchscript_onnx_qnn: - inference_time: 946.0 - throughput: 1057.0824524312895 + inference_time: 941.0 + throughput: 1062.6992561105208 estimated_peak_memory_range: - min: 184320 - max: 1666632 + min: 180224 + max: 1451960 primary_compute_unit: NPU precision: int8 layer_info: @@ -181,7 +181,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jqp4eervg + job_id: j1glq0llp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -190,13 +190,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:55:39Z' + timestamp: '2024-08-27T00:11:57Z' - torchscript_onnx_tflite: - inference_time: 939.0 - throughput: 1064.9627263045793 + inference_time: 910.0 + throughput: 1098.901098901099 estimated_peak_memory_range: min: 16384 - max: 64173536 + max: 64968384 primary_compute_unit: NPU precision: int8 layer_info: @@ -204,14 +204,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jz5wylnjg + job_id: j0pxz7x95 job_status: Passed torchscript_onnx_qnn: - inference_time: 1160.0 - throughput: 862.0689655172414 + inference_time: 1132.0 + throughput: 883.3922261484099 estimated_peak_memory_range: - min: 184320 - max: 16598064 + min: 167936 + max: 20376368 primary_compute_unit: NPU precision: int8 layer_info: @@ -219,7 +219,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: joprxx39p + job_id: j1pvn17mg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -228,13 +228,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:55:44Z' + timestamp: '2024-08-27T00:12:01Z' - torchscript_onnx_tflite: - inference_time: 797.0 - throughput: 1254.7051442910915 + inference_time: 781.0 + throughput: 1280.4097311139565 estimated_peak_memory_range: - min: 12288 - max: 1345016 + min: 20480 + max: 13733360 primary_compute_unit: NPU precision: int8 layer_info: @@ -242,14 +242,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jmg9ozevg + job_id: jo5mlw8qg job_status: Passed torchscript_onnx_qnn: - inference_time: 943.0 - throughput: 1060.4453870625662 + inference_time: 948.0 + throughput: 1054.8523206751054 estimated_peak_memory_range: - min: 176128 - max: 1489592 + min: 204800 + max: 1883800 primary_compute_unit: NPU precision: int8 layer_info: @@ -257,7 +257,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j0px00o1p + job_id: jw5603w75 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -266,13 +266,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:55:41Z' + timestamp: '2024-08-27T00:11:58Z' - torchscript_onnx_tflite: - inference_time: 808.0 - throughput: 1237.6237623762377 + inference_time: 783.0 + throughput: 1277.139208173691 estimated_peak_memory_range: min: 24576 - max: 271321608 + max: 229621016 primary_compute_unit: NPU precision: int8 layer_info: @@ -280,14 +280,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jnp1onxl5 + job_id: jegnw9kmg job_status: Passed torchscript_onnx_qnn: inference_time: 944.0 throughput: 1059.322033898305 estimated_peak_memory_range: min: 180224 - max: 1938528 + max: 1389200 primary_compute_unit: NPU precision: int8 layer_info: @@ -295,7 +295,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jo5m99xwg + job_id: j1p3r46zp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -304,13 +304,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:55:42Z' + timestamp: '2024-08-27T00:11:59Z' - torchscript_onnx_tflite: - inference_time: 787.0 - throughput: 1270.6480304955528 + inference_time: 779.0 + throughput: 1283.6970474967907 estimated_peak_memory_range: min: 12288 - max: 46713368 + max: 57643896 primary_compute_unit: NPU precision: int8 layer_info: @@ -318,14 +318,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jvgd6dllp + job_id: jopr74weg job_status: Passed torchscript_onnx_qnn: - inference_time: 952.0 - throughput: 1050.420168067227 + inference_time: 954.0 + throughput: 1048.2180293501049 estimated_peak_memory_range: - min: 184320 - max: 1415936 + min: 180224 + max: 1533808 primary_compute_unit: NPU precision: int8 layer_info: @@ -333,7 +333,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jegn11vrp + job_id: jwgo918dg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -342,13 +342,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:55:43Z' + timestamp: '2024-08-27T00:12:00Z' - torchscript_onnx_tflite: - inference_time: 2763.0 - throughput: 361.92544335866813 + inference_time: 2688.0 + throughput: 372.0238095238095 estimated_peak_memory_range: min: 12288 - max: 25226464 + max: 26610832 primary_compute_unit: NPU precision: int8 layer_info: @@ -356,14 +356,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jz5wyy66g + job_id: jqpyy4m4p job_status: Passed torchscript_onnx_qnn: - inference_time: 3987.0 - throughput: 250.8151492350138 + inference_time: 4133.0 + throughput: 241.95499637067505 estimated_peak_memory_range: - min: 208896 - max: 8391792 + min: 163840 + max: 8098560 primary_compute_unit: NPU precision: int8 layer_info: @@ -371,7 +371,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jep2ooy4g + job_id: j7gj80q85 job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -380,13 +380,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:55:46Z' + timestamp: '2024-08-27T00:12:02Z' - torchscript_onnx_tflite: - inference_time: 11526.0 - throughput: 86.76036786395974 + inference_time: 11395.0 + throughput: 87.7577885037297 estimated_peak_memory_range: - min: 40960 - max: 2211664 + min: 20480 + max: 7080136 primary_compute_unit: NPU precision: int8 layer_info: @@ -394,7 +394,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jmg9oonlg + job_id: j2p0x16ep job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -403,13 +403,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:55:35Z' + timestamp: '2024-08-27T00:11:53Z' - torchscript_onnx_qnn: - inference_time: 1002.0 - throughput: 998.003992015968 + inference_time: 1127.0 + throughput: 887.3114463176574 estimated_peak_memory_range: - min: 458752 - max: 458752 + min: 1605632 + max: 1605632 primary_compute_unit: NPU precision: int8 layer_info: @@ -417,14 +417,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jz57oorlg + job_id: jn5qd7vmg job_status: Passed torchscript_onnx: - inference_time: 1068.0 - throughput: 936.3295880149813 + inference_time: 1011.0 + throughput: 989.1196834817013 estimated_peak_memory_range: - min: 30171136 - max: 30171136 + min: 30052352 + max: 30052352 primary_compute_unit: NPU precision: int8 layer_info: @@ -432,7 +432,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: j1p8jjyx5 + job_id: jz5wrd4jp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -441,4 +441,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:55:49Z' + timestamp: '2024-08-27T00:12:05Z' diff --git a/qai_hub_models/models/resnext101/perf.yaml b/qai_hub_models/models/resnext101/perf.yaml index d5a7ae0a..49777085 100644 --- a/qai_hub_models/models/resnext101/perf.yaml +++ b/qai_hub_models/models/resnext101/perf.yaml @@ -45,11 +45,11 @@ models: - name: ResNeXt101 performance_metrics: - torchscript_onnx_tflite: - inference_time: 6455.0 - throughput: 154.91866769945779 + inference_time: 6758.0 + throughput: 147.9727730097662 estimated_peak_memory_range: - min: 45056 - max: 2298568 + min: 36864 + max: 2289736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jygzzl96g + job_id: jygz04mz5 job_status: Passed torchscript_onnx_qnn: - inference_time: 6681.0 - throughput: 149.678191887442 + inference_time: 6798.0 + throughput: 147.10208884966167 estimated_peak_memory_range: min: 16384 - max: 35218864 + max: 35094888 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: j0px0l29p + job_id: j0pxzd4j5 job_status: Passed torchscript_onnx: - inference_time: 6801.0 - throughput: 147.03720041170416 + inference_time: 7038.0 + throughput: 142.08581983518044 estimated_peak_memory_range: - min: 49152 - max: 203205224 + min: 12288 + max: 203118344 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 247 - job_id: jogk62mo5 + job_id: jogkk92yg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:54:25Z' + timestamp: '2024-08-27T11:51:01Z' - torchscript_onnx_tflite: - inference_time: 4774.0 - throughput: 209.46795140343528 + inference_time: 4611.0 + throughput: 216.8726957276079 estimated_peak_memory_range: - min: 20480 - max: 369083392 + min: 40960 + max: 366920384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jz5wylvjg + job_id: jz5wr17zp job_status: Passed torchscript_onnx_qnn: - inference_time: 4835.0 - throughput: 206.82523267838675 + inference_time: 4761.0 + throughput: 210.03990758244066 estimated_peak_memory_range: min: 618496 - max: 83783424 + max: 83195008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jo5m90yqg + job_id: jo5mldmyg job_status: Passed torchscript_onnx: - inference_time: 4874.0 - throughput: 205.1702913418137 + inference_time: 4836.0 + throughput: 206.782464846981 estimated_peak_memory_range: min: 0 - max: 368847392 + max: 367452192 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 247 - job_id: jn5q4lomg + job_id: jn5qdml7g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:54:26Z' + timestamp: '2024-08-27T11:51:02Z' - torchscript_onnx_tflite: - inference_time: 6493.0 - throughput: 154.01201293700908 + inference_time: 6463.0 + throughput: 154.7269070091289 estimated_peak_memory_range: - min: 65536 - max: 26335800 + min: 32768 + max: 2073720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jmg9oz1vg + job_id: jmg9qxmqp job_status: Passed torchscript_onnx_qnn: - inference_time: 6675.0 - throughput: 149.81273408239701 + inference_time: 6757.0 + throughput: 147.9946721918011 estimated_peak_memory_range: - min: 679936 - max: 2003128 + min: 643072 + max: 1858800 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: joprxljep + job_id: jopr7nlvg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:54:19Z' + timestamp: '2024-08-27T11:50:57Z' - torchscript_onnx_tflite: - inference_time: 9213.0 - throughput: 108.54227721697602 + inference_time: 9189.0 + throughput: 108.82576994232234 estimated_peak_memory_range: min: 20480 - max: 161062608 + max: 160970848 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jnp1onll5 + job_id: jnp1mvjkp job_status: Passed torchscript_onnx_qnn: - inference_time: 9428.0 - throughput: 106.06703436571914 + inference_time: 9312.0 + throughput: 107.38831615120274 estimated_peak_memory_range: - min: 638976 - max: 51308928 + min: 618496 + max: 49501056 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: j1p8jev85 + job_id: j1p8k4ezp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:54:23Z' + timestamp: '2024-08-27T11:51:01Z' - torchscript_onnx_tflite: - inference_time: 6702.0 - throughput: 149.20919128618323 + inference_time: 6519.0 + throughput: 153.39776039269827 estimated_peak_memory_range: - min: 24576 - max: 2335344 + min: 36864 + max: 2460232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jvgd6d9lp + job_id: jvgdmz3kg job_status: Passed torchscript_onnx_qnn: - inference_time: 6753.0 - throughput: 148.08233377758035 + inference_time: 6778.0 + throughput: 147.5361463558572 estimated_peak_memory_range: - min: 2113536 - max: 3296944 + min: 659456 + max: 1918432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jep2ornmg + job_id: jep2zvrxp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:54:20Z' + timestamp: '2024-08-27T11:50:58Z' - torchscript_onnx_tflite: - inference_time: 6480.0 - throughput: 154.320987654321 + inference_time: 6523.0 + throughput: 153.30369461904033 estimated_peak_memory_range: - min: 24576 - max: 2381456 + min: 45056 + max: 2264776 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jz57oe3rg + job_id: jz57874qp job_status: Passed torchscript_onnx_qnn: - inference_time: 6608.0 - throughput: 151.3317191283293 + inference_time: 6733.0 + throughput: 148.5222040695084 estimated_peak_memory_range: - min: 626688 - max: 1813904 + min: 634880 + max: 1895984 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jqpy8o04g + job_id: jqpyy7orp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:54:21Z' + timestamp: '2024-08-27T11:50:59Z' - torchscript_onnx_tflite: - inference_time: 6491.0 - throughput: 154.05946695424433 + inference_time: 6467.0 + throughput: 154.63120457708365 estimated_peak_memory_range: - min: 24576 - max: 1647984 + min: 36864 + max: 2380368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 147 - job_id: jqp4ey0lg + job_id: jqp4291qg job_status: Passed torchscript_onnx_qnn: - inference_time: 6761.0 - throughput: 147.90711433219937 + inference_time: 6881.0 + throughput: 145.32771399505887 estimated_peak_memory_range: - min: 626688 - max: 2220160 + min: 638976 + max: 1815280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: j2p0om7ep + job_id: j2p0xvm2p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:54:22Z' + timestamp: '2024-08-27T11:51:00Z' - torchscript_onnx_qnn: - inference_time: 6872.0 - throughput: 145.51804423748544 + inference_time: 6833.0 + throughput: 146.34860237084735 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 245 - job_id: jegn1z8mp + job_id: jegnw7zvg job_status: Passed torchscript_onnx: - inference_time: 6846.0 - throughput: 146.0706982179375 + inference_time: 6741.0 + throughput: 148.3459427384661 estimated_peak_memory_range: - min: 182435840 - max: 182435840 + min: 182673408 + max: 182673408 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 247 - job_id: j1glwyrlp + job_id: j1glq1yep job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:54:27Z' + timestamp: '2024-08-27T11:51:03Z' diff --git a/qai_hub_models/models/resnext101_quantized/perf.yaml b/qai_hub_models/models/resnext101_quantized/perf.yaml index eec3dea8..ce3be684 100644 --- a/qai_hub_models/models/resnext101_quantized/perf.yaml +++ b/qai_hub_models/models/resnext101_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: ResNeXt101Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 2770.0 - throughput: 361.01083032490976 + inference_time: 2768.0 + throughput: 361.271676300578 estimated_peak_memory_range: min: 16384 - max: 2035840 + max: 2688880 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jlpe6790g + job_id: jlpenly7p job_status: Passed torchscript_onnx_qnn: - inference_time: 3034.0 - throughput: 329.5978905735003 + inference_time: 3027.0 + throughput: 330.3600925008259 estimated_peak_memory_range: - min: 12288 - max: 33645104 + min: 16384 + max: 33216008 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jo5m902qg + job_id: jo5mld8yg job_status: Passed torchscript_onnx: - inference_time: 3237.0 - throughput: 308.9280197713933 + inference_time: 3313.0 + throughput: 301.84123151222457 estimated_peak_memory_range: - min: 69632 - max: 2786440 + min: 16384 + max: 2684064 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 151 - job_id: j1glwyzlp + job_id: j1glq1xep job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:53:37Z' + timestamp: '2024-08-27T11:50:20Z' - torchscript_onnx_tflite: - inference_time: 2117.0 - throughput: 472.3665564478035 + inference_time: 2090.0 + throughput: 478.4688995215311 estimated_peak_memory_range: min: 12288 - max: 265135744 + max: 271609760 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jygzzle6g + job_id: jygz04nz5 job_status: Passed torchscript_onnx_qnn: - inference_time: 2359.0 - throughput: 423.908435777872 + inference_time: 2304.0 + throughput: 434.02777777777777 estimated_peak_memory_range: - min: 0 - max: 82814576 + min: 12775424 + max: 98133328 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jegn1zymp + job_id: jegnw7nvg job_status: Passed torchscript_onnx: - inference_time: 2492.0 - throughput: 401.2841091492777 + inference_time: 2451.0 + throughput: 407.9967360261118 estimated_peak_memory_range: min: 0 - max: 292281856 + max: 301223600 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 151 - job_id: jw56o8j75 + job_id: jw560d7v5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:53:38Z' + timestamp: '2024-08-27T11:50:21Z' - torchscript_onnx_tflite: - inference_time: 2751.0 - throughput: 363.50418029807344 + inference_time: 2762.0 + throughput: 362.0564808110065 estimated_peak_memory_range: min: 12288 - max: 1827648 + max: 2082040 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jz5wylojg + job_id: jz5wr14zp job_status: Passed torchscript_onnx_qnn: - inference_time: 2939.0 - throughput: 340.25178632187817 + inference_time: 2940.0 + throughput: 340.13605442176873 estimated_peak_memory_range: - min: 184320 - max: 1508728 + min: 188416 + max: 1748904 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jep2or6mg + job_id: jep2zvwxp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:53:29Z' + timestamp: '2024-08-27T11:50:15Z' - torchscript_onnx_tflite: - inference_time: 3370.0 - throughput: 296.7359050445104 + inference_time: 3369.0 + throughput: 296.8239833778569 estimated_peak_memory_range: - min: 16384 - max: 265356560 + min: 20480 + max: 272798800 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jmg9ozvvg + job_id: jmg9qxdqp job_status: Passed torchscript_onnx_qnn: - inference_time: 3659.0 - throughput: 273.2987154960372 + inference_time: 3623.0 + throughput: 276.0143527463428 estimated_peak_memory_range: min: 12288 - max: 80825184 + max: 93772256 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jogk62no5 + job_id: jogkk94yg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:53:34Z' + timestamp: '2024-08-27T11:50:19Z' - torchscript_onnx_tflite: - inference_time: 2741.0 - throughput: 364.8303538854433 + inference_time: 2814.0 + throughput: 355.36602700781805 estimated_peak_memory_range: - min: 24576 - max: 1766632 + min: 20480 + max: 2695328 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jnp1on0l5 + job_id: jnp1mv6kp job_status: Passed torchscript_onnx_qnn: - inference_time: 2904.0 - throughput: 344.3526170798898 + inference_time: 2870.0 + throughput: 348.4320557491289 estimated_peak_memory_range: - min: 196608 - max: 1637224 + min: 188416 + max: 1785528 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jqpy8ow4g + job_id: jqpyy7xrp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:53:30Z' + timestamp: '2024-08-27T11:50:16Z' - torchscript_onnx_tflite: - inference_time: 2745.0 - throughput: 364.29872495446267 + inference_time: 2830.0 + throughput: 353.35689045936397 estimated_peak_memory_range: - min: 16384 - max: 2281264 + min: 24576 + max: 2201752 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jvgd6dwlp + job_id: jvgdmz2kg job_status: Passed torchscript_onnx_qnn: - inference_time: 2980.0 - throughput: 335.5704697986577 + inference_time: 3002.0 + throughput: 333.11125916055965 estimated_peak_memory_range: - min: 118784 - max: 1414680 + min: 176128 + max: 1306040 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: j2p0omqep + job_id: j2p0xvj2p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:53:32Z' + timestamp: '2024-08-27T11:50:17Z' - torchscript_onnx_tflite: - inference_time: 2736.0 - throughput: 365.4970760233918 + inference_time: 2754.0 + throughput: 363.10820624546113 estimated_peak_memory_range: min: 16384 - max: 38786888 + max: 15730448 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jz57oewrg + job_id: jz57879qp job_status: Passed torchscript_onnx_qnn: - inference_time: 2921.0 - throughput: 342.3485107839781 + inference_time: 2968.0 + throughput: 336.92722371967653 estimated_peak_memory_range: min: 180224 - max: 1676728 + max: 1735352 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: j1p8je985 + job_id: j1p8k4xzp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:53:33Z' + timestamp: '2024-08-27T11:50:18Z' - torchscript_onnx_tflite: - inference_time: 9956.0 - throughput: 100.44194455604661 + inference_time: 9992.0 + throughput: 100.080064051241 estimated_peak_memory_range: min: 12288 - max: 200253712 + max: 208892784 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 150 - job_id: jqp4eyolg + job_id: jqp4293qg job_status: Passed torchscript_onnx_qnn: - inference_time: 14524.0 - throughput: 68.85155604516662 + inference_time: 14597.0 + throughput: 68.50722751250257 estimated_peak_memory_range: - min: 188416 - max: 8607840 + min: 172032 + max: 7546256 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: jn5q4lkmg + job_id: jn5qdmy7g job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:53:36Z' + timestamp: '2024-08-27T11:50:20Z' - torchscript_onnx_tflite: - inference_time: 140932.0 - throughput: 7.095620582976187 + inference_time: 110179.0 + throughput: 9.076139736247379 estimated_peak_memory_range: - min: 4960256 - max: 559577424 + min: 4845568 + max: 361896760 primary_compute_unit: GPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 125 layers_on_cpu: 11 total_layers: 150 - job_id: j0px0lj9p + job_id: j0pxzdxj5 job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:53:25Z' + timestamp: '2024-08-27T11:50:11Z' - torchscript_onnx_qnn: - inference_time: 3189.0 - throughput: 313.5779241141424 + inference_time: 3033.0 + throughput: 329.7065611605671 estimated_peak_memory_range: - min: 196608 - max: 196608 + min: 208896 + max: 208896 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 146 - job_id: joprxlqep + job_id: jopr7n0vg job_status: Passed torchscript_onnx: - inference_time: 3193.0 - throughput: 313.18509238960223 + inference_time: 3173.0 + throughput: 315.1591553734636 estimated_peak_memory_range: - min: 95727616 - max: 95727616 + min: 95191040 + max: 95191040 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 151 - job_id: j1p3oz3zp + job_id: j1p3rw9xp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:53:39Z' + timestamp: '2024-08-27T11:50:22Z' diff --git a/qai_hub_models/models/resnext50/perf.yaml b/qai_hub_models/models/resnext50/perf.yaml index 95bae656..c87d8ffe 100644 --- a/qai_hub_models/models/resnext50/perf.yaml +++ b/qai_hub_models/models/resnext50/perf.yaml @@ -45,11 +45,11 @@ models: - name: ResNeXt50 performance_metrics: - torchscript_onnx_tflite: - inference_time: 2559.0 - throughput: 390.77764751856193 + inference_time: 2539.0 + throughput: 393.8558487593541 estimated_peak_memory_range: - min: 65536 - max: 2524896 + min: 16384 + max: 1975024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jz5wyljjg + job_id: jqp42xjqg job_status: Passed torchscript_onnx_qnn: - inference_time: 2623.0 - throughput: 381.2428516965307 + inference_time: 2589.0 + throughput: 386.24951718810354 estimated_peak_memory_range: - min: 57344 - max: 34400448 + min: 401408 + max: 30435264 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jo5m90rqg + job_id: j2p0x1k2p job_status: Passed torchscript_onnx: - inference_time: 2746.0 - throughput: 364.1660597232338 + inference_time: 2752.0 + throughput: 363.3720930232558 estimated_peak_memory_range: - min: 524288 - max: 2973416 + min: 12288 + max: 60140792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jn5q4l8mg + job_id: j1pvn147g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:51:32Z' + timestamp: '2024-08-27T00:08:02Z' - torchscript_onnx_tflite: - inference_time: 1760.0 - throughput: 568.1818181818181 + inference_time: 1776.0 + throughput: 563.063063063063 estimated_peak_memory_range: min: 12288 - max: 177503328 + max: 177943472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jmg9oz6vg + job_id: j0pxz7ej5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1855.0 - throughput: 539.0835579514825 + inference_time: 1839.0 + throughput: 543.773790103317 estimated_peak_memory_range: min: 618496 - max: 38033472 + max: 35238528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jegn1z2mp + job_id: j1p8k38zp job_status: Passed torchscript_onnx: - inference_time: 1966.0 - throughput: 508.646998982706 + inference_time: 1932.0 + throughput: 517.5983436853002 estimated_peak_memory_range: min: 0 - max: 178202592 + max: 178072880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: j1glwynlp + job_id: j7gj80175 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:51:33Z' + timestamp: '2024-08-27T00:08:03Z' - torchscript_onnx_tflite: - inference_time: 2489.0 - throughput: 401.76777822418643 + inference_time: 2486.0 + throughput: 402.2526146419952 estimated_peak_memory_range: - min: 0 - max: 1698672 + min: 16384 + max: 61717792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jnp1onrl5 + job_id: jo5mlwvyg job_status: Passed torchscript_onnx_qnn: - inference_time: 2506.0 - throughput: 399.0422984836393 + inference_time: 2483.0 + throughput: 402.7386226339106 estimated_peak_memory_range: - min: 634880 - max: 1947752 + min: 638976 + max: 1958976 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jep2or8mg + job_id: jn5qd7w7g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:51:26Z' + timestamp: '2024-08-27T00:07:56Z' - torchscript_onnx_tflite: - inference_time: 3298.0 - throughput: 303.21406913280777 + inference_time: 3251.0 + throughput: 307.59766225776684 estimated_peak_memory_range: min: 16384 - max: 114414864 + max: 114665168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jvgd6djlp + job_id: jegnw9xvg job_status: Passed torchscript_onnx_qnn: - inference_time: 3345.0 - throughput: 298.9536621823617 + inference_time: 3393.0 + throughput: 294.7244326554671 estimated_peak_memory_range: min: 618496 - max: 27080896 + max: 26608432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jogk62zo5 + job_id: jwgo91m4g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:51:31Z' + timestamp: '2024-08-27T00:08:01Z' - torchscript_onnx_tflite: - inference_time: 2466.0 - throughput: 405.51500405515003 + inference_time: 2469.0 + throughput: 405.0222762251924 estimated_peak_memory_range: - min: 45056 - max: 2162616 + min: 184320 + max: 1968024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jz57oezrg + job_id: jopr749vg job_status: Passed torchscript_onnx_qnn: - inference_time: 2499.0 - throughput: 400.16006402561027 + inference_time: 2475.0 + throughput: 404.04040404040404 estimated_peak_memory_range: - min: 634880 - max: 2106952 + min: 626688 + max: 1796088 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jqpy8oe4g + job_id: j1glq07ep job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:51:27Z' + timestamp: '2024-08-27T00:07:57Z' - torchscript_onnx_tflite: - inference_time: 2477.0 - throughput: 403.7141703673799 + inference_time: 2516.0 + throughput: 397.456279809221 estimated_peak_memory_range: - min: 28672 - max: 2420368 + min: 65536 + max: 2202464 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jqp4eyqlg + job_id: jep2z7jxp job_status: Passed torchscript_onnx_qnn: - inference_time: 2504.0 - throughput: 399.36102236421726 + inference_time: 2473.0 + throughput: 404.36716538617065 estimated_peak_memory_range: - min: 651264 - max: 2031048 + min: 634880 + max: 2075792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j2p0omyep + job_id: jw5603vv5 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:51:28Z' + timestamp: '2024-08-27T00:07:58Z' - torchscript_onnx_tflite: - inference_time: 2478.0 - throughput: 403.5512510088781 + inference_time: 2460.0 + throughput: 406.5040650406504 estimated_peak_memory_range: - min: 24576 - max: 2424128 + min: 0 + max: 1785192 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: j0px0lv9p + job_id: jqpyy4nrp job_status: Passed torchscript_onnx_qnn: - inference_time: 2521.0 - throughput: 396.6679888932963 + inference_time: 2527.0 + throughput: 395.7261574990107 estimated_peak_memory_range: min: 634880 - max: 1840640 + max: 2312728 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1p8jeo85 + job_id: j1p3r48xp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:51:30Z' + timestamp: '2024-08-27T00:08:00Z' - torchscript_onnx_qnn: - inference_time: 2651.0 - throughput: 377.2161448509996 + inference_time: 2634.0 + throughput: 379.65072133637057 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: joprxlkep + job_id: jogkkldyg job_status: Passed torchscript_onnx: - inference_time: 2670.0 - throughput: 374.53183520599254 + inference_time: 2749.0 + throughput: 363.7686431429611 estimated_peak_memory_range: - min: 53972992 - max: 53972992 + min: 54046720 + max: 54046720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jw56o8675 + job_id: jlpenr27p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:51:34Z' + timestamp: '2024-08-27T00:08:04Z' diff --git a/qai_hub_models/models/resnext50_quantized/perf.yaml b/qai_hub_models/models/resnext50_quantized/perf.yaml index 369f40fd..44f862a3 100644 --- a/qai_hub_models/models/resnext50_quantized/perf.yaml +++ b/qai_hub_models/models/resnext50_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: ResNeXt50Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 920.0 - throughput: 1086.9565217391305 + inference_time: 915.0 + throughput: 1092.896174863388 estimated_peak_memory_range: min: 12288 - max: 284236312 + max: 1432376 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jvgd6dkkp + job_id: jnp1mddkp job_status: Passed torchscript_onnx_qnn: - inference_time: 1167.0 - throughput: 856.898029134533 + inference_time: 1169.0 + throughput: 855.4319931565441 estimated_peak_memory_range: - min: 12288 - max: 22983184 + min: 172032 + max: 11683392 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jegn1zjmp + job_id: jqpyy4vrp job_status: Passed torchscript_onnx: - inference_time: 1282.0 - throughput: 780.0312012480499 + inference_time: 1333.0 + throughput: 750.1875468867216 estimated_peak_memory_range: - min: 16384 - max: 31444760 + min: 12288 + max: 30870552 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: jw56o8k75 + job_id: j7gj80k75 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:50:51Z' + timestamp: '2024-08-27T00:07:20Z' - torchscript_onnx_tflite: - inference_time: 698.0 - throughput: 1432.6647564469913 + inference_time: 686.0 + throughput: 1457.725947521866 estimated_peak_memory_range: min: 12288 - max: 102449056 + max: 105061648 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jz5wylkjg + job_id: jvgdmrrkg job_status: Passed torchscript_onnx_qnn: - inference_time: 878.0 - throughput: 1138.9521640091116 + inference_time: 876.0 + throughput: 1141.552511415525 estimated_peak_memory_range: - min: 167936 - max: 33207712 + min: 163840 + max: 34870352 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: joprxlzep + job_id: j2p0x1e2p job_status: Passed torchscript_onnx: - inference_time: 1014.0 - throughput: 986.1932938856016 + inference_time: 991.0 + throughput: 1009.0817356205853 estimated_peak_memory_range: min: 28672 - max: 122807376 + max: 125975392 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: j1p3ozyzp + job_id: jlpenr47p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:50:52Z' + timestamp: '2024-08-27T00:07:21Z' - torchscript_onnx_tflite: - inference_time: 933.0 - throughput: 1071.8113612004288 + inference_time: 908.0 + throughput: 1101.3215859030836 estimated_peak_memory_range: - min: 12288 - max: 13181616 + min: 28672 + max: 318920656 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jmg9ozrvg + job_id: jz578jjqp job_status: Passed torchscript_onnx_qnn: - inference_time: 1096.0 - throughput: 912.4087591240876 + inference_time: 1115.0 + throughput: 896.8609865470852 estimated_peak_memory_range: - min: 176128 - max: 1364552 + min: 184320 + max: 1399440 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jqpy8o94g + job_id: jogkklryg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:50:44Z' + timestamp: '2024-08-27T00:07:15Z' - torchscript_onnx_tflite: - inference_time: 1122.0 - throughput: 891.2655971479501 + inference_time: 1089.0 + throughput: 918.2736455463728 estimated_peak_memory_range: - min: 16384 - max: 103917456 + min: 12288 + max: 108562272 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jnp1on9l5 + job_id: jqp42xxqg job_status: Passed torchscript_onnx_qnn: - inference_time: 1369.0 - throughput: 730.4601899196493 + inference_time: 1345.0 + throughput: 743.4944237918215 estimated_peak_memory_range: min: 167936 - max: 36047728 + max: 37560704 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jn5q4ljmg + job_id: jwgo91e4g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:50:49Z' + timestamp: '2024-08-27T00:07:19Z' - torchscript_onnx_tflite: - inference_time: 918.0 - throughput: 1089.3246187363834 + inference_time: 904.0 + throughput: 1106.1946902654868 estimated_peak_memory_range: min: 12288 - max: 20692472 + max: 25718296 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jvgd6dklp + job_id: j0pxz77j5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1098.0 - throughput: 910.7468123861566 + inference_time: 1124.0 + throughput: 889.6797153024911 estimated_peak_memory_range: min: 184320 - max: 1427536 + max: 1495224 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j2p0omnep + job_id: jn5qd797g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:50:45Z' + timestamp: '2024-08-27T00:07:16Z' - torchscript_onnx_tflite: - inference_time: 918.0 - throughput: 1089.3246187363834 + inference_time: 923.0 + throughput: 1083.4236186348862 estimated_peak_memory_range: - min: 24576 - max: 323611536 + min: 12288 + max: 8880624 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jz57oeqrg + job_id: jo5mlwwyg job_status: Passed torchscript_onnx_qnn: - inference_time: 1097.0 - throughput: 911.5770282588878 + inference_time: 1113.0 + throughput: 898.4725965858041 estimated_peak_memory_range: min: 184320 - max: 1677152 + max: 1413008 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j1p8jel85 + job_id: j1glq0eep job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:50:46Z' + timestamp: '2024-08-27T00:07:17Z' - torchscript_onnx_tflite: - inference_time: 918.0 - throughput: 1089.3246187363834 + inference_time: 921.0 + throughput: 1085.7763300760043 estimated_peak_memory_range: - min: 28672 - max: 11624128 + min: 12288 + max: 22880344 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jqp4eyzlg + job_id: jegnw9rvg job_status: Passed torchscript_onnx_qnn: - inference_time: 1120.0 - throughput: 892.8571428571429 + inference_time: 1122.0 + throughput: 891.2655971479501 estimated_peak_memory_range: min: 184320 - max: 1394928 + max: 1638104 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jogk62jo5 + job_id: jw5603qv5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:50:47Z' + timestamp: '2024-08-27T00:07:18Z' - torchscript_onnx_tflite: - inference_time: 3038.0 - throughput: 329.1639236339697 + inference_time: 2998.0 + throughput: 333.55570380253505 estimated_peak_memory_range: min: 12288 - max: 57784032 + max: 60515824 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: j0px0lw9p + job_id: jopr741vg job_status: Passed torchscript_onnx_qnn: - inference_time: 4514.0 - throughput: 221.53300841825433 + inference_time: 4448.0 + throughput: 224.82014388489208 estimated_peak_memory_range: min: 212992 - max: 8919248 + max: 8114048 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j1glwyjlp + job_id: j1pvn1z7g job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:50:50Z' + timestamp: '2024-08-27T00:07:20Z' - torchscript_onnx_tflite: - inference_time: 62784.0 - throughput: 15.927624872579 + inference_time: 55358.0 + throughput: 18.064236424726328 estimated_peak_memory_range: - min: 8327168 - max: 25144912 + min: 0 + max: 97061304 primary_compute_unit: GPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 57 layers_on_cpu: 11 total_layers: 82 - job_id: jo5m90jqg + job_id: jep2z73xp job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:50:39Z' + timestamp: '2024-08-27T00:07:11Z' - torchscript_onnx_qnn: - inference_time: 1205.0 - throughput: 829.8755186721992 + inference_time: 1409.0 + throughput: 709.7232079488999 estimated_peak_memory_range: - min: 499712 - max: 499712 + min: 1597440 + max: 1597440 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jep2or2mg + job_id: j1p8k3wzp job_status: Passed torchscript_onnx: - inference_time: 1369.0 - throughput: 730.4601899196493 + inference_time: 1291.0 + throughput: 774.5933384972889 estimated_peak_memory_range: - min: 30724096 - max: 30724096 + min: 30711808 + max: 30711808 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: jwgodljd5 + job_id: jygz0xvz5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:50:53Z' + timestamp: '2024-08-27T00:07:22Z' diff --git a/qai_hub_models/models/sam/export.py b/qai_hub_models/models/sam/export.py index 6df4b5eb..ec28bc9b 100644 --- a/qai_hub_models/models/sam/export.py +++ b/qai_hub_models/models/sam/export.py @@ -129,8 +129,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/sam/perf.yaml b/qai_hub_models/models/sam/perf.yaml index 07c116cf..83cda0ef 100644 --- a/qai_hub_models/models/sam/perf.yaml +++ b/qai_hub_models/models/sam/perf.yaml @@ -43,11 +43,11 @@ models: - name: SAMDecoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 47928.0 - throughput: 20.86463027875146 + inference_time: 30062.0 + throughput: 33.26458652118954 estimated_peak_memory_range: - min: 4284416 - max: 22687768 + min: 4317184 + max: 23238504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -55,7 +55,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 337 - job_id: jvgd6dykp + job_id: jnp1md8kp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -64,13 +64,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:49:13Z' + timestamp: '2024-08-27T00:05:50Z' - torchscript_onnx_tflite: - inference_time: 33893.0 - throughput: 29.504617472634468 + inference_time: 21034.0 + throughput: 47.54207473614149 estimated_peak_memory_range: - min: 1388544 - max: 223942320 + min: 2183168 + max: 215251840 primary_compute_unit: NPU precision: fp16 layer_info: @@ -78,7 +78,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 337 - job_id: jqp4ey6qg + job_id: jz578jdqp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -87,13 +87,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:49:15Z' + timestamp: '2024-08-27T00:05:52Z' - torchscript_onnx_tflite: - inference_time: 47748.0 - throughput: 20.943285582642204 + inference_time: 30130.0 + throughput: 33.18951211417192 estimated_peak_memory_range: - min: 4239360 - max: 323343752 + min: 4026368 + max: 23036784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -101,7 +101,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 337 - job_id: jo5m901yg + job_id: j0pxz71j5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -110,13 +110,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:49:18Z' + timestamp: '2024-08-27T00:05:53Z' - torchscript_onnx_tflite: - inference_time: 52190.0 - throughput: 19.160758766047135 + inference_time: 33349.0 + throughput: 29.985906623886773 estimated_peak_memory_range: - min: 4087808 - max: 218164464 + min: 12288 + max: 202875280 primary_compute_unit: NPU precision: fp16 layer_info: @@ -124,7 +124,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 337 - job_id: joprxlrvp + job_id: jegnw99vg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -133,13 +133,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:49:20Z' + timestamp: '2024-08-27T00:05:55Z' - torchscript_onnx_tflite: - inference_time: 47871.0 - throughput: 20.889473794155126 + inference_time: 30298.0 + throughput: 33.005478909498976 estimated_peak_memory_range: - min: 4063232 - max: 22177120 + min: 4022272 + max: 12271000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -147,7 +147,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 337 - job_id: jqpy8olrg + job_id: jep2z77xp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -156,13 +156,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:49:22Z' + timestamp: '2024-08-27T00:05:56Z' - torchscript_onnx_tflite: - inference_time: 48467.0 - throughput: 20.632595374172116 + inference_time: 30333.0 + throughput: 32.9673952461016 estimated_peak_memory_range: - min: 4939776 - max: 23203232 + min: 3997696 + max: 23191160 primary_compute_unit: NPU precision: fp16 layer_info: @@ -170,7 +170,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 337 - job_id: j1p8jenz5 + job_id: j2p0x112p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -179,13 +179,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:49:24Z' + timestamp: '2024-08-27T00:05:58Z' - torchscript_onnx_tflite: - inference_time: 48421.0 - throughput: 20.652196361083 + inference_time: 30183.0 + throughput: 33.13123281317298 estimated_peak_memory_range: - min: 2207744 - max: 20913160 + min: 4018176 + max: 22964504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -193,7 +193,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 337 - job_id: jn5q4ln7g + job_id: jogkkllyg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -202,15 +202,15 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:49:26Z' + timestamp: '2024-08-27T00:06:00Z' - name: SAMEncoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 10523061.0 - throughput: 0.09502938356054384 + inference_time: 11442656.0 + throughput: 0.08739229773227475 estimated_peak_memory_range: - min: 2603032576 - max: 2650450584 + min: 2715643904 + max: 2722451832 primary_compute_unit: CPU precision: fp32 layer_info: @@ -218,7 +218,7 @@ models: layers_on_gpu: 36 layers_on_cpu: 782 total_layers: 818 - job_id: jz57oe1qg + job_id: jvgdmrvkg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -227,13 +227,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:49:14Z' + timestamp: '2024-08-27T00:05:50Z' - torchscript_onnx_tflite: - inference_time: 9544271.0 - throughput: 0.10477489585113416 + inference_time: 9743957.0 + throughput: 0.10262771069289407 estimated_peak_memory_range: - min: 2566725632 - max: 3483704080 + min: 2562260992 + max: 4090295936 primary_compute_unit: CPU precision: fp32 layer_info: @@ -241,7 +241,7 @@ models: layers_on_gpu: 36 layers_on_cpu: 782 total_layers: 818 - job_id: j0px0l8jp + job_id: jqp42xwqg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -250,13 +250,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:49:16Z' + timestamp: '2024-08-27T00:05:52Z' - torchscript_onnx_tflite: - inference_time: 11206076.0 - throughput: 0.08923730304881031 + inference_time: 11160798.0 + throughput: 0.08959932793336103 estimated_peak_memory_range: - min: 2718027776 - max: 2721555480 + min: 2711654400 + max: 2722142984 primary_compute_unit: CPU precision: fp32 layer_info: @@ -264,7 +264,7 @@ models: layers_on_gpu: 36 layers_on_cpu: 782 total_layers: 818 - job_id: jegn1z4vp + job_id: jo5mlwzyg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -273,13 +273,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:49:18Z' + timestamp: '2024-08-27T00:05:53Z' - torchscript_onnx_tflite: - inference_time: 14590800.0 - throughput: 0.06853633796638978 + inference_time: 16328120.0 + throughput: 0.06124403789291113 estimated_peak_memory_range: - min: 2556588032 - max: 3435794688 + min: 2498056192 + max: 4088972800 primary_compute_unit: CPU precision: fp32 layer_info: @@ -287,7 +287,7 @@ models: layers_on_gpu: 36 layers_on_cpu: 782 total_layers: 818 - job_id: jep2or1xg + job_id: jopr744vg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -296,13 +296,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:49:20Z' + timestamp: '2024-08-27T00:05:55Z' - torchscript_onnx_tflite: - inference_time: 10561919.0 - throughput: 0.09467976416028186 + inference_time: 11097886.0 + throughput: 0.09010725105664268 estimated_peak_memory_range: - min: 2713022464 - max: 2721451696 + min: 2713260032 + max: 2721063416 primary_compute_unit: CPU precision: fp32 layer_info: @@ -310,7 +310,7 @@ models: layers_on_gpu: 36 layers_on_cpu: 782 total_layers: 818 - job_id: j2p0omw2p + job_id: jqpyy44rp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -319,13 +319,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:49:22Z' + timestamp: '2024-08-27T00:05:57Z' - torchscript_onnx_tflite: - inference_time: 11874414.0 - throughput: 0.08421468208873296 + inference_time: 10821911.0 + throughput: 0.09240512142448778 estimated_peak_memory_range: - min: 2648469504 - max: 2652692768 + min: 2708471808 + max: 2716254896 primary_compute_unit: CPU precision: fp32 layer_info: @@ -333,7 +333,7 @@ models: layers_on_gpu: 36 layers_on_cpu: 782 total_layers: 818 - job_id: jogk621y5 + job_id: j1p8k33zp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -342,13 +342,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:49:24Z' + timestamp: '2024-08-27T00:05:58Z' - torchscript_onnx_tflite: - inference_time: 11251672.0 - throughput: 0.08887567998782758 + inference_time: 11208498.0 + throughput: 0.08921802011295359 estimated_peak_memory_range: - min: 2717110272 - max: 2720499696 + min: 2653159424 + max: 2717096464 primary_compute_unit: CPU precision: fp32 layer_info: @@ -356,7 +356,7 @@ models: layers_on_gpu: 36 layers_on_cpu: 782 total_layers: 818 - job_id: j1glwydep + job_id: jn5qd777g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -365,4 +365,4 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:49:26Z' + timestamp: '2024-08-27T00:06:00Z' diff --git a/qai_hub_models/models/sesr_m5/perf.yaml b/qai_hub_models/models/sesr_m5/perf.yaml index 62bf262e..0ddd7cd1 100644 --- a/qai_hub_models/models/sesr_m5/perf.yaml +++ b/qai_hub_models/models/sesr_m5/perf.yaml @@ -45,11 +45,11 @@ models: - name: SESR-M5 performance_metrics: - torchscript_onnx_tflite: - inference_time: 2304.0 - throughput: 434.02777777777777 + inference_time: 2335.0 + throughput: 428.2655246252677 estimated_peak_memory_range: - min: 24576 - max: 2218928 + min: 12668928 + max: 14075856 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 25 - job_id: jqp4ey2qg + job_id: jo5mlw37g job_status: Passed torchscript_onnx_qnn: - inference_time: 2150.0 - throughput: 465.1162790697674 + inference_time: 2130.0 + throughput: 469.4835680751174 estimated_peak_memory_range: - min: 16384 - max: 3585864 + min: 212992 + max: 4791384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: j2p0om82p + job_id: jogkklyvg job_status: Passed torchscript_onnx: - inference_time: 2924.0 - throughput: 341.9972640218878 + inference_time: 2909.0 + throughput: 343.7607425232039 estimated_peak_memory_range: - min: 12288 - max: 2786528 + min: 16384 + max: 30094968 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 33 - job_id: j1pv2ly7g + job_id: jlpenrv8p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:48:08Z' + timestamp: '2024-08-27T00:04:46Z' - torchscript_onnx_tflite: - inference_time: 1553.0 - throughput: 643.915003219575 + inference_time: 1509.0 + throughput: 662.6905235255136 estimated_peak_memory_range: min: 0 - max: 26878416 + max: 27682320 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 25 - job_id: j0px0lzjp + job_id: jegnw93jg job_status: Passed torchscript_onnx_qnn: - inference_time: 1470.0 - throughput: 680.2721088435375 + inference_time: 1463.0 + throughput: 683.526999316473 estimated_peak_memory_range: - min: 208896 - max: 13050528 + min: 204800 + max: 13296928 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: j1p8jedz5 + job_id: jn5qd72eg job_status: Passed torchscript_onnx: - inference_time: 1914.0 - throughput: 522.466039707419 + inference_time: 1903.0 + throughput: 525.4860746190226 estimated_peak_memory_range: min: 0 - max: 30301360 + max: 29966000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 33 - job_id: j7gj3r67p + job_id: jygz0x745 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:48:09Z' + timestamp: '2024-08-27T00:04:47Z' - torchscript_onnx_tflite: - inference_time: 2270.0 - throughput: 440.52863436123346 + inference_time: 2233.0 + throughput: 447.82803403493057 estimated_peak_memory_range: min: 16384 - max: 9609520 + max: 4317384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 25 - job_id: jo5m90lyg + job_id: jopr74ekg job_status: Passed torchscript_onnx_qnn: - inference_time: 2248.0 - throughput: 444.83985765124555 + inference_time: 2141.0 + throughput: 467.07146193367583 estimated_peak_memory_range: - min: 221184 - max: 1521408 + min: 229376 + max: 1951136 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jn5q4lx7g + job_id: jw56031n5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:48:02Z' + timestamp: '2024-08-27T00:04:41Z' - torchscript_onnx_tflite: - inference_time: 3397.0 - throughput: 294.3773918163085 + inference_time: 4504.0 + throughput: 222.02486678507992 estimated_peak_memory_range: - min: 12603392 - max: 40628240 + min: 12607488 + max: 40636672 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 25 - job_id: jegn1zdvp + job_id: jep2z7l6p job_status: Passed torchscript_onnx_qnn: - inference_time: 3163.0 - throughput: 316.1555485298767 + inference_time: 3184.0 + throughput: 314.07035175879395 estimated_peak_memory_range: - min: 204800 - max: 14990896 + min: 208896 + max: 16787808 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jwgodl745 + job_id: j7gj80l15 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:48:07Z' + timestamp: '2024-08-27T00:04:45Z' - torchscript_onnx_tflite: - inference_time: 2247.0 - throughput: 445.0378282153983 + inference_time: 2275.0 + throughput: 439.56043956043953 estimated_peak_memory_range: min: 16384 - max: 25251192 + max: 72047360 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 25 - job_id: joprxlmvp + job_id: jqpyy460p job_status: Passed torchscript_onnx_qnn: - inference_time: 2249.0 - throughput: 444.642063139173 + inference_time: 2157.0 + throughput: 463.60686138154847 estimated_peak_memory_range: min: 229376 - max: 1455256 + max: 1326912 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: j1glwy9ep + job_id: j1p3r4mmp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:48:03Z' + timestamp: '2024-08-27T00:04:42Z' - torchscript_onnx_tflite: - inference_time: 2310.0 - throughput: 432.9004329004329 + inference_time: 2208.0 + throughput: 452.8985507246377 estimated_peak_memory_range: min: 28672 - max: 1431840 + max: 1418496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 25 - job_id: jep2orqxg + job_id: j2p0x1r0p job_status: Passed torchscript_onnx_qnn: - inference_time: 2249.0 - throughput: 444.642063139173 + inference_time: 2146.0 + throughput: 465.98322460391427 estimated_peak_memory_range: - min: 221184 - max: 1881400 + min: 237568 + max: 1833352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jw56o89v5 + job_id: jwgo91v1g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:48:04Z' + timestamp: '2024-08-27T00:04:43Z' - torchscript_onnx_tflite: - inference_time: 2257.0 - throughput: 443.06601683650865 + inference_time: 2183.0 + throughput: 458.0852038479157 estimated_peak_memory_range: - min: 20480 - max: 1604744 + min: 24576 + max: 5826816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 25 - job_id: jqpy8okrg + job_id: j1p8k37qp job_status: Passed torchscript_onnx_qnn: - inference_time: 2412.0 - throughput: 414.5936981757877 + inference_time: 2539.0 + throughput: 393.8558487593541 estimated_peak_memory_range: - min: 225280 - max: 1406056 + min: 229376 + max: 1683888 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: j1p3ozlxp + job_id: j1pvn1wzg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:48:06Z' + timestamp: '2024-08-27T00:04:44Z' - torchscript_onnx_qnn: - inference_time: 2591.0 - throughput: 385.95137012736393 + inference_time: 2328.0 + throughput: 429.553264604811 estimated_peak_memory_range: - min: 208896 - max: 208896 + min: 212992 + max: 212992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 31 - job_id: jogk62wy5 + job_id: j1glq0k2p job_status: Passed torchscript_onnx: - inference_time: 3073.0 - throughput: 325.4149040026033 + inference_time: 2961.0 + throughput: 337.7237419790611 estimated_peak_memory_range: - min: 8990720 - max: 8990720 + min: 8978432 + max: 8978432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 33 - job_id: jlpe6707g + job_id: jz5wrd94p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:48:10Z' + timestamp: '2024-08-27T00:04:48Z' diff --git a/qai_hub_models/models/sesr_m5_quantized/perf.yaml b/qai_hub_models/models/sesr_m5_quantized/perf.yaml index 28331599..435b808e 100644 --- a/qai_hub_models/models/sesr_m5_quantized/perf.yaml +++ b/qai_hub_models/models/sesr_m5_quantized/perf.yaml @@ -48,11 +48,11 @@ models: - name: SESR-M5-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 1407.0 - throughput: 710.7320540156361 + inference_time: 1333.0 + throughput: 750.1875468867216 estimated_peak_memory_range: - min: 20480 - max: 16105624 + min: 32768 + max: 80486544 primary_compute_unit: NPU precision: int8 layer_info: @@ -60,14 +60,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: jz57oenqg + job_id: jqp42xd2g job_status: Passed torchscript_onnx_qnn: - inference_time: 972.0 - throughput: 1028.80658436214 + inference_time: 984.0 + throughput: 1016.260162601626 estimated_peak_memory_range: - min: 12288 - max: 3683744 + min: 28672 + max: 4056800 primary_compute_unit: NPU precision: int8 layer_info: @@ -75,14 +75,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: j1p8jekz5 + job_id: jogkkl3vg job_status: Passed torchscript_onnx: - inference_time: 1055.0 - throughput: 947.8672985781991 + inference_time: 1066.0 + throughput: 938.0863039399625 estimated_peak_memory_range: min: 12288 - max: 2066048 + max: 1865112 primary_compute_unit: NPU precision: int8 layer_info: @@ -90,7 +90,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 29 - job_id: jlpe67n7g + job_id: jygz0xr45 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -99,13 +99,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:47:30Z' + timestamp: '2024-08-27T00:04:12Z' - torchscript_onnx_tflite: - inference_time: 1154.0 - throughput: 866.5511265164645 + inference_time: 1122.0 + throughput: 891.2655971479501 estimated_peak_memory_range: - min: 0 - max: 25026448 + min: 12288 + max: 25853776 primary_compute_unit: NPU precision: int8 layer_info: @@ -113,14 +113,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: jqp4ey4qg + job_id: j0pxz7685 job_status: Passed torchscript_onnx_qnn: - inference_time: 708.0 - throughput: 1412.4293785310736 + inference_time: 705.0 + throughput: 1418.4397163120568 estimated_peak_memory_range: - min: 65536 - max: 13486160 + min: 61440 + max: 15552752 primary_compute_unit: NPU precision: int8 layer_info: @@ -128,14 +128,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: jogk62ky5 + job_id: jn5qd73eg job_status: Passed torchscript_onnx: - inference_time: 797.0 - throughput: 1254.7051442910915 + inference_time: 799.0 + throughput: 1251.5644555694619 estimated_peak_memory_range: min: 0 - max: 27054000 + max: 28304192 primary_compute_unit: NPU precision: int8 layer_info: @@ -143,7 +143,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 29 - job_id: jygzzl0zg + job_id: jz5wrdq4p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -152,13 +152,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:47:31Z' + timestamp: '2024-08-27T00:04:13Z' - torchscript_onnx_tflite: - inference_time: 1437.0 - throughput: 695.8942240779402 + inference_time: 1332.0 + throughput: 750.7507507507507 estimated_peak_memory_range: - min: 12288 - max: 1333024 + min: 20480 + max: 81475880 primary_compute_unit: NPU precision: int8 layer_info: @@ -166,14 +166,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: j0px0lrjp + job_id: jo5mlw67g job_status: Passed torchscript_onnx_qnn: - inference_time: 761.0 - throughput: 1314.060446780552 + inference_time: 770.0 + throughput: 1298.7012987012988 estimated_peak_memory_range: - min: 81920 - max: 2324360 + min: 24576 + max: 2161672 primary_compute_unit: NPU precision: int8 layer_info: @@ -181,7 +181,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: j1glwyqep + job_id: jw5603nn5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -190,13 +190,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:47:23Z' + timestamp: '2024-08-27T00:04:06Z' - torchscript_onnx_tflite: - inference_time: 2652.0 - throughput: 377.0739064856712 + inference_time: 1760.0 + throughput: 568.1818181818181 estimated_peak_memory_range: - min: 3166208 - max: 28358800 + min: 12288 + max: 25900976 primary_compute_unit: NPU precision: int8 layer_info: @@ -204,14 +204,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: jo5m90kyg + job_id: jegnw9mjg job_status: Passed torchscript_onnx_qnn: - inference_time: 1107.0 - throughput: 903.342366757001 + inference_time: 1110.0 + throughput: 900.9009009009009 estimated_peak_memory_range: - min: 61440 - max: 12873904 + min: 65536 + max: 16192352 primary_compute_unit: NPU precision: int8 layer_info: @@ -219,7 +219,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: j1pv2ln7g + job_id: j7gj80e15 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -228,13 +228,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:47:28Z' + timestamp: '2024-08-27T00:04:10Z' - torchscript_onnx_tflite: - inference_time: 1390.0 - throughput: 719.4244604316547 + inference_time: 1323.0 + throughput: 755.8578987150415 estimated_peak_memory_range: - min: 24576 - max: 1612224 + min: 12288 + max: 1397648 primary_compute_unit: NPU precision: int8 layer_info: @@ -242,14 +242,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: jegn1zwvp + job_id: jopr742kg job_status: Passed torchscript_onnx_qnn: - inference_time: 771.0 - throughput: 1297.0168612191958 + inference_time: 767.0 + throughput: 1303.7809647979138 estimated_peak_memory_range: - min: 86016 - max: 1406552 + min: 102400 + max: 1303888 primary_compute_unit: NPU precision: int8 layer_info: @@ -257,7 +257,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: jw56o80v5 + job_id: j1p3r4emp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -266,13 +266,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:47:24Z' + timestamp: '2024-08-27T00:04:07Z' - torchscript_onnx_tflite: - inference_time: 1382.0 - throughput: 723.589001447178 + inference_time: 1326.0 + throughput: 754.1478129713424 estimated_peak_memory_range: - min: 28672 - max: 1477056 + min: 12288 + max: 1428320 primary_compute_unit: NPU precision: int8 layer_info: @@ -280,14 +280,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: joprxl7vp + job_id: jep2z796p job_status: Passed torchscript_onnx_qnn: - inference_time: 766.0 - throughput: 1305.4830287206266 + inference_time: 763.0 + throughput: 1310.615989515072 estimated_peak_memory_range: - min: 81920 - max: 1509296 + min: 102400 + max: 1227664 primary_compute_unit: NPU precision: int8 layer_info: @@ -295,7 +295,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: j1p3ozrxp + job_id: jwgo9131g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -304,13 +304,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:47:25Z' + timestamp: '2024-08-27T00:04:08Z' - torchscript_onnx_tflite: - inference_time: 1390.0 - throughput: 719.4244604316547 + inference_time: 1332.0 + throughput: 750.7507507507507 estimated_peak_memory_range: - min: 24576 - max: 1379848 + min: 20480 + max: 1442496 primary_compute_unit: NPU precision: int8 layer_info: @@ -318,14 +318,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: jep2orzxg + job_id: jqpyy4j0p job_status: Passed torchscript_onnx_qnn: - inference_time: 767.0 - throughput: 1303.7809647979138 + inference_time: 769.0 + throughput: 1300.3901170351105 estimated_peak_memory_range: - min: 24576 - max: 1839800 + min: 77824 + max: 1501672 primary_compute_unit: NPU precision: int8 layer_info: @@ -333,7 +333,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: jwgodl945 + job_id: j1pvn1vzg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -342,13 +342,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:47:27Z' + timestamp: '2024-08-27T00:04:09Z' - torchscript_onnx_tflite: - inference_time: 3895.0 - throughput: 256.73940949935815 + inference_time: 4101.0 + throughput: 243.84296513045598 estimated_peak_memory_range: - min: 3203072 - max: 20946512 + min: 3211264 + max: 22550224 primary_compute_unit: NPU precision: int8 layer_info: @@ -356,14 +356,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: jqpy8oyrg + job_id: j2p0x1l0p job_status: Passed torchscript_onnx_qnn: - inference_time: 2974.0 - throughput: 336.2474781439139 + inference_time: 2999.0 + throughput: 333.4444814938313 estimated_peak_memory_range: - min: 61440 - max: 8232976 + min: 12288 + max: 8432128 primary_compute_unit: NPU precision: int8 layer_info: @@ -371,7 +371,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: j7gj3r87p + job_id: jlpenrk8p job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -380,13 +380,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:47:29Z' + timestamp: '2024-08-27T00:04:11Z' - torchscript_onnx_tflite: - inference_time: 21031.0 - throughput: 47.548856450002376 + inference_time: 18089.0 + throughput: 55.28221571120571 estimated_peak_memory_range: - min: 3354624 - max: 10069664 + min: 3424256 + max: 5055440 primary_compute_unit: NPU precision: int8 layer_info: @@ -394,7 +394,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 27 - job_id: j2p0omx2p + job_id: j1p8k3zqp job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -403,13 +403,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:47:18Z' + timestamp: '2024-08-27T00:04:02Z' - torchscript_onnx_qnn: - inference_time: 881.0 - throughput: 1135.0737797956867 + inference_time: 905.0 + throughput: 1104.9723756906078 estimated_peak_memory_range: - min: 65536 - max: 65536 + min: 1101824 + max: 1101824 primary_compute_unit: NPU precision: int8 layer_info: @@ -417,14 +417,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 25 - job_id: jn5q4ld7g + job_id: j1glq032p job_status: Passed torchscript_onnx: - inference_time: 1097.0 - throughput: 911.5770282588878 + inference_time: 1121.0 + throughput: 892.0606601248885 estimated_peak_memory_range: - min: 3448832 - max: 3448832 + min: 3854336 + max: 3854336 primary_compute_unit: NPU precision: int8 layer_info: @@ -432,7 +432,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 29 - job_id: jz5wylrzg + job_id: jmg9q3wmp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -441,4 +441,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:47:32Z' + timestamp: '2024-08-27T00:04:13Z' diff --git a/qai_hub_models/models/shufflenet_v2/perf.yaml b/qai_hub_models/models/shufflenet_v2/perf.yaml index 5f681d0b..ce0f904d 100644 --- a/qai_hub_models/models/shufflenet_v2/perf.yaml +++ b/qai_hub_models/models/shufflenet_v2/perf.yaml @@ -45,11 +45,11 @@ models: - name: Shufflenet-v2 performance_metrics: - torchscript_onnx_tflite: - inference_time: 1236.0 - throughput: 809.0614886731391 + inference_time: 1201.0 + throughput: 832.6394671107411 estimated_peak_memory_range: - min: 20480 - max: 114009880 + min: 12288 + max: 11322000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 204 - job_id: j0px0l0jp + job_id: j0pxz7n85 job_status: Passed torchscript_onnx_qnn: - inference_time: 778.0 - throughput: 1285.3470437017995 + inference_time: 771.0 + throughput: 1297.0168612191958 estimated_peak_memory_range: - min: 16384 - max: 103862000 + min: 622592 + max: 96427464 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: j1p8jerz5 + job_id: j1p8k3mqp job_status: Passed torchscript_onnx: - inference_time: 1059.0 - throughput: 944.2870632672333 + inference_time: 1080.0 + throughput: 925.925925925926 estimated_peak_memory_range: - min: 12288 - max: 4573384 + min: 20480 + max: 4732912 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: j7gj3ry7p + job_id: j7gj80215 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:46:47Z' + timestamp: '2024-08-27T00:03:33Z' - torchscript_onnx_tflite: - inference_time: 788.0 - throughput: 1269.0355329949239 + inference_time: 777.0 + throughput: 1287.001287001287 estimated_peak_memory_range: min: 12288 - max: 38239472 + max: 38218592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 204 - job_id: jo5m909yg + job_id: jo5mlwq7g job_status: Passed torchscript_onnx_qnn: - inference_time: 510.0 - throughput: 1960.7843137254902 + inference_time: 515.0 + throughput: 1941.7475728155339 estimated_peak_memory_range: - min: 0 - max: 13768928 + min: 618496 + max: 12949136 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jogk620y5 + job_id: jogkklqvg job_status: Passed torchscript_onnx: - inference_time: 727.0 - throughput: 1375.515818431912 + inference_time: 745.0 + throughput: 1342.2818791946308 estimated_peak_memory_range: min: 0 - max: 40463680 + max: 40944880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: jlpe67x7g + job_id: jlpenrw8p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:46:48Z' + timestamp: '2024-08-27T00:03:34Z' - torchscript_onnx_tflite: - inference_time: 1237.0 - throughput: 808.4074373484236 + inference_time: 1195.0 + throughput: 836.8200836820083 estimated_peak_memory_range: - min: 20480 - max: 1432496 + min: 16384 + max: 12261320 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 204 - job_id: jegn1zqvp + job_id: jegnw9ljg job_status: Passed torchscript_onnx_qnn: - inference_time: 763.0 - throughput: 1310.615989515072 + inference_time: 764.0 + throughput: 1308.9005235602094 estimated_peak_memory_range: - min: 0 - max: 1197600 + min: 643072 + max: 2240216 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: j1glwy8ep + job_id: j1glq022p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:46:41Z' + timestamp: '2024-08-27T00:03:29Z' - torchscript_onnx_tflite: - inference_time: 1354.0 - throughput: 738.5524372230428 + inference_time: 1327.0 + throughput: 753.5795026375282 estimated_peak_memory_range: min: 12288 - max: 39867248 + max: 39890736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 204 - job_id: joprxldvp + job_id: jopr748kg job_status: Passed torchscript_onnx_qnn: - inference_time: 879.0 - throughput: 1137.6564277588168 + inference_time: 871.0 + throughput: 1148.105625717566 estimated_peak_memory_range: - min: 618496 - max: 13640832 + min: 622592 + max: 13868416 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: j1pv2lm7g + job_id: j1pvn1rzg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:46:46Z' + timestamp: '2024-08-27T00:03:32Z' - torchscript_onnx_tflite: - inference_time: 1228.0 - throughput: 814.3322475570033 + inference_time: 1196.0 + throughput: 836.1204013377926 estimated_peak_memory_range: - min: 36864 - max: 1387824 + min: 28672 + max: 6746528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 204 - job_id: jep2ordxg + job_id: jep2z706p job_status: Passed torchscript_onnx_qnn: - inference_time: 769.0 - throughput: 1300.3901170351105 + inference_time: 757.0 + throughput: 1321.003963011889 estimated_peak_memory_range: - min: 622592 - max: 1943792 + min: 634880 + max: 2236920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jw56o8mv5 + job_id: jw5603zn5 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:46:42Z' + timestamp: '2024-08-27T00:03:29Z' - torchscript_onnx_tflite: - inference_time: 1236.0 - throughput: 809.0614886731391 + inference_time: 1201.0 + throughput: 832.6394671107411 estimated_peak_memory_range: min: 12288 - max: 24185816 + max: 1510552 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 204 - job_id: jqpy8o2rg + job_id: jqpyy4r0p job_status: Passed torchscript_onnx_qnn: - inference_time: 761.0 - throughput: 1314.060446780552 + inference_time: 760.0 + throughput: 1315.7894736842106 estimated_peak_memory_range: - min: 630784 - max: 1892704 + min: 647168 + max: 1911784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: j1p3oz7xp + job_id: j1p3r41mp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:46:44Z' + timestamp: '2024-08-27T00:03:30Z' - torchscript_onnx_tflite: - inference_time: 1239.0 - throughput: 807.1025020177563 + inference_time: 1197.0 + throughput: 835.421888053467 estimated_peak_memory_range: min: 20480 - max: 113781936 + max: 6192576 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 204 - job_id: j2p0om92p + job_id: j2p0x120p job_status: Passed torchscript_onnx_qnn: - inference_time: 763.0 - throughput: 1310.615989515072 + inference_time: 760.0 + throughput: 1315.7894736842106 estimated_peak_memory_range: - min: 630784 - max: 2141864 + min: 622592 + max: 1976432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jwgodlw45 + job_id: jwgo91n1g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:46:45Z' + timestamp: '2024-08-27T00:03:31Z' - torchscript_onnx_qnn: - inference_time: 909.0 - throughput: 1100.1100110011 + inference_time: 904.0 + throughput: 1106.1946902654868 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 158 - job_id: jn5q4l17g + job_id: jn5qd7reg job_status: Passed torchscript_onnx: - inference_time: 1136.0 - throughput: 880.2816901408451 + inference_time: 1189.0 + throughput: 841.0428931875525 estimated_peak_memory_range: - min: 3989504 - max: 3989504 + min: 5505024 + max: 5505024 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 223 - job_id: jygzzlyzg + job_id: jygz0xj45 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:46:49Z' + timestamp: '2024-08-27T00:03:35Z' diff --git a/qai_hub_models/models/shufflenet_v2_quantized/perf.yaml b/qai_hub_models/models/shufflenet_v2_quantized/perf.yaml index 72b59a5a..3f346d3f 100644 --- a/qai_hub_models/models/shufflenet_v2_quantized/perf.yaml +++ b/qai_hub_models/models/shufflenet_v2_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: Shufflenet-v2Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 633.0 - throughput: 1579.778830963665 + inference_time: 606.0 + throughput: 1650.1650165016501 estimated_peak_memory_range: min: 12288 - max: 75622688 + max: 65699784 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 207 - job_id: jep2orr6g + job_id: j0pxz7985 job_status: Passed torchscript_onnx_qnn: - inference_time: 591.0 - throughput: 1692.047377326565 + inference_time: 586.0 + throughput: 1706.4846416382252 estimated_peak_memory_range: - min: 12288 - max: 66191296 + min: 167936 + max: 3701176 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,7 +78,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: jwgodld15 + job_id: jn5qd7eeg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -87,13 +87,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:45:59Z' + timestamp: '2024-08-27T00:02:51Z' - torchscript_onnx_tflite: - inference_time: 430.0 - throughput: 2325.5813953488373 + inference_time: 429.0 + throughput: 2331.002331002331 estimated_peak_memory_range: min: 12288 - max: 25389472 + max: 26944448 primary_compute_unit: NPU precision: int8 layer_info: @@ -101,14 +101,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 207 - job_id: jqpy8oo0g + job_id: jo5mlwe7g job_status: Passed torchscript_onnx_qnn: - inference_time: 421.0 - throughput: 2375.296912114014 + inference_time: 422.0 + throughput: 2369.6682464454975 estimated_peak_memory_range: - min: 0 - max: 14687296 + min: 159744 + max: 13551968 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,7 +116,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: j1pv2l2zg + job_id: j1glq062p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -125,13 +125,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:46:00Z' + timestamp: '2024-08-27T00:02:52Z' - torchscript_onnx_tflite: - inference_time: 618.0 - throughput: 1618.1229773462783 + inference_time: 605.0 + throughput: 1652.892561983471 estimated_peak_memory_range: min: 12288 - max: 1263488 + max: 1506984 primary_compute_unit: NPU precision: int8 layer_info: @@ -139,14 +139,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 207 - job_id: j2p0omo0p + job_id: jegnw90jg job_status: Passed torchscript_onnx_qnn: - inference_time: 546.0 - throughput: 1831.5018315018315 + inference_time: 547.0 + throughput: 1828.1535648994516 estimated_peak_memory_range: - min: 176128 - max: 1506672 + min: 172032 + max: 1439552 primary_compute_unit: NPU precision: int8 layer_info: @@ -154,7 +154,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: jlpe6768g + job_id: j1p3r4vmp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -163,13 +163,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:46:02Z' + timestamp: '2024-08-27T00:02:54Z' - torchscript_onnx_tflite: - inference_time: 670.0 - throughput: 1492.5373134328358 + inference_time: 648.0 + throughput: 1543.20987654321 estimated_peak_memory_range: - min: 16384 - max: 26700064 + min: 12288 + max: 28051680 primary_compute_unit: NPU precision: int8 layer_info: @@ -177,14 +177,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 207 - job_id: j1p8jejq5 + job_id: jopr746kg job_status: Passed torchscript_onnx_qnn: - inference_time: 628.0 - throughput: 1592.3566878980891 + inference_time: 632.0 + throughput: 1582.2784810126582 estimated_peak_memory_range: min: 159744 - max: 14095760 + max: 15495616 primary_compute_unit: NPU precision: int8 layer_info: @@ -192,7 +192,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: jnp1onon5 + job_id: jlpenre8p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -201,13 +201,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:46:07Z' + timestamp: '2024-08-27T00:02:58Z' - torchscript_onnx_tflite: - inference_time: 624.0 - throughput: 1602.5641025641025 + inference_time: 610.0 + throughput: 1639.344262295082 estimated_peak_memory_range: min: 12288 - max: 1388736 + max: 4654008 primary_compute_unit: NPU precision: int8 layer_info: @@ -215,14 +215,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 207 - job_id: jogk626v5 + job_id: jep2z7x6p job_status: Passed torchscript_onnx_qnn: - inference_time: 547.0 - throughput: 1828.1535648994516 + inference_time: 543.0 + throughput: 1841.6206261510129 estimated_peak_memory_range: - min: 176128 - max: 1611936 + min: 184320 + max: 1630376 primary_compute_unit: NPU precision: int8 layer_info: @@ -230,7 +230,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: jygzzlz4g + job_id: jwgo91k1g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -239,13 +239,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:46:04Z' + timestamp: '2024-08-27T00:02:55Z' - torchscript_onnx_tflite: - inference_time: 635.0 - throughput: 1574.8031496062993 + inference_time: 610.0 + throughput: 1639.344262295082 estimated_peak_memory_range: min: 12288 - max: 75254048 + max: 1504880 primary_compute_unit: NPU precision: int8 layer_info: @@ -253,14 +253,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 207 - job_id: jn5q4l4eg + job_id: jqpyy4z0p job_status: Passed torchscript_onnx_qnn: - inference_time: 546.0 - throughput: 1831.5018315018315 + inference_time: 542.0 + throughput: 1845.018450184502 estimated_peak_memory_range: - min: 184320 - max: 1610728 + min: 176128 + max: 1448352 primary_compute_unit: NPU precision: int8 layer_info: @@ -268,7 +268,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: jz5wyly4g + job_id: j1pvn10zg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -277,13 +277,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:46:05Z' + timestamp: '2024-08-27T00:02:56Z' - torchscript_onnx_tflite: - inference_time: 626.0 - throughput: 1597.444089456869 + inference_time: 610.0 + throughput: 1639.344262295082 estimated_peak_memory_range: min: 12288 - max: 75264016 + max: 1760776 primary_compute_unit: NPU precision: int8 layer_info: @@ -291,14 +291,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 207 - job_id: j1glwyw2p + job_id: j2p0x130p job_status: Passed torchscript_onnx_qnn: - inference_time: 544.0 - throughput: 1838.235294117647 + inference_time: 541.0 + throughput: 1848.4288354898335 estimated_peak_memory_range: - min: 180224 - max: 1413800 + min: 176128 + max: 1423080 primary_compute_unit: NPU precision: int8 layer_info: @@ -306,7 +306,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: jmg9ozomg + job_id: j7gj80z15 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -315,13 +315,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:46:06Z' + timestamp: '2024-08-27T00:02:57Z' - torchscript_onnx_tflite: - inference_time: 954.0 - throughput: 1048.2180293501049 + inference_time: 942.0 + throughput: 1061.5711252653928 estimated_peak_memory_range: - min: 16384 - max: 20525264 + min: 12288 + max: 20703888 primary_compute_unit: NPU precision: int8 layer_info: @@ -329,14 +329,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 207 - job_id: jw56o8on5 + job_id: j1p8k30qp job_status: Passed torchscript_onnx_qnn: - inference_time: 1103.0 - throughput: 906.6183136899365 + inference_time: 1109.0 + throughput: 901.7132551848512 estimated_peak_memory_range: min: 12288 - max: 7988080 + max: 7496048 primary_compute_unit: NPU precision: int8 layer_info: @@ -344,7 +344,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: jvgd6d66p + job_id: jygz0xo45 job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -353,13 +353,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:46:08Z' + timestamp: '2024-08-27T00:02:59Z' - torchscript_onnx_tflite: - inference_time: 10762.0 - throughput: 92.9195316855603 + inference_time: 9060.0 + throughput: 110.37527593818984 estimated_peak_memory_range: - min: 176128 - max: 5763344 + min: 36864 + max: 5756168 primary_compute_unit: CPU precision: fp32 layer_info: @@ -367,7 +367,7 @@ models: layers_on_gpu: 9 layers_on_cpu: 154 total_layers: 207 - job_id: j1p3ozomp + job_id: jogkkl7vg job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -376,13 +376,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:45:57Z' + timestamp: '2024-08-27T00:02:50Z' - torchscript_onnx_qnn: - inference_time: 660.0 - throughput: 1515.1515151515152 + inference_time: 763.0 + throughput: 1310.615989515072 estimated_peak_memory_range: - min: 499712 - max: 499712 + min: 532480 + max: 532480 primary_compute_unit: NPU precision: int8 layer_info: @@ -390,7 +390,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 122 - job_id: j7gj3r31p + job_id: jw5603en5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -399,4 +399,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:46:01Z' + timestamp: '2024-08-27T00:02:52Z' diff --git a/qai_hub_models/models/sinet/perf.yaml b/qai_hub_models/models/sinet/perf.yaml index 2e7a6589..44cef89a 100644 --- a/qai_hub_models/models/sinet/perf.yaml +++ b/qai_hub_models/models/sinet/perf.yaml @@ -45,11 +45,11 @@ models: - name: SINet performance_metrics: - torchscript_onnx_tflite: - inference_time: 1759.0 - throughput: 568.5048322910744 + inference_time: 1734.0 + throughput: 576.7012687427913 estimated_peak_memory_range: - min: 12288 - max: 1517736 + min: 438272 + max: 2030616 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 240 - job_id: j2p0omm0p + job_id: j2p0x1znp job_status: Passed torchscript_onnx_qnn: - inference_time: 1195.0 - throughput: 836.8200836820083 + inference_time: 1185.0 + throughput: 843.8818565400844 estimated_peak_memory_range: - min: 618496 - max: 75549624 + min: 36864 + max: 77353648 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jwgodll15 + job_id: jwgo91zkg job_status: Passed torchscript_onnx: - inference_time: 2270.0 - throughput: 440.52863436123346 + inference_time: 2269.0 + throughput: 440.72278536800354 estimated_peak_memory_range: - min: 622592 - max: 2527976 + min: 618496 + max: 2399672 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 229 - job_id: jvgd6dd6p + job_id: jvgdmrnzg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:45:16Z' + timestamp: '2024-08-27T00:02:12Z' - torchscript_onnx_tflite: - inference_time: 1137.0 - throughput: 879.5074758135444 + inference_time: 1129.0 + throughput: 885.7395925597874 estimated_peak_memory_range: min: 16384 - max: 31031184 + max: 31329264 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 240 - job_id: j1p8jeeq5 + job_id: j1p8k3qop job_status: Passed torchscript_onnx_qnn: - inference_time: 789.0 - throughput: 1267.427122940431 + inference_time: 792.0 + throughput: 1262.6262626262626 estimated_peak_memory_range: min: 0 - max: 14337776 + max: 14400016 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: j1pv2llzg + job_id: j1pvn1qrg job_status: Passed torchscript_onnx: - inference_time: 1566.0 - throughput: 638.5696040868455 + inference_time: 1542.0 + throughput: 648.5084306095979 estimated_peak_memory_range: min: 0 - max: 34099040 + max: 34824656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 229 - job_id: jz57oeeng + job_id: jz5wrdw4p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:45:17Z' + timestamp: '2024-08-27T00:02:13Z' - torchscript_onnx_tflite: - inference_time: 1758.0 - throughput: 568.8282138794084 + inference_time: 1728.0 + throughput: 578.7037037037037 estimated_peak_memory_range: - min: 12288 - max: 1277008 + min: 24576 + max: 1453848 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 240 - job_id: jogk622v5 + job_id: jogkkleng job_status: Passed torchscript_onnx_qnn: - inference_time: 1126.0 - throughput: 888.0994671403197 + inference_time: 1149.0 + throughput: 870.3220191470845 estimated_peak_memory_range: - min: 634880 - max: 2239888 + min: 630784 + max: 2182120 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jlpe6778g + job_id: jlpenrovp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:45:10Z' + timestamp: '2024-08-27T00:02:07Z' - torchscript_onnx_tflite: - inference_time: 1915.0 - throughput: 522.1932114882507 + inference_time: 1890.0 + throughput: 529.1005291005291 estimated_peak_memory_range: min: 12288 - max: 31369360 + max: 30874736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 240 - job_id: jn5q4lleg + job_id: jn5qd76og job_status: Passed torchscript_onnx_qnn: - inference_time: 1329.0 - throughput: 752.4454477050414 + inference_time: 1317.0 + throughput: 759.3014426727411 estimated_peak_memory_range: min: 618496 - max: 15858688 + max: 17453520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jnp1onnn5 + job_id: jnp1md27p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:45:15Z' + timestamp: '2024-08-27T00:02:11Z' - torchscript_onnx_tflite: - inference_time: 1763.0 - throughput: 567.2149744753261 + inference_time: 1736.0 + throughput: 576.036866359447 estimated_peak_memory_range: - min: 32768 - max: 1367280 + min: 12288 + max: 1883864 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 240 - job_id: j1glwyy2p + job_id: j1glq04mp job_status: Passed torchscript_onnx_qnn: - inference_time: 1134.0 - throughput: 881.8342151675485 + inference_time: 1171.0 + throughput: 853.9709649871904 estimated_peak_memory_range: - min: 630784 - max: 1877696 + min: 634880 + max: 1988728 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jygzzll4g + job_id: jygz0x2x5 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:45:12Z' + timestamp: '2024-08-27T00:02:08Z' - torchscript_onnx_tflite: - inference_time: 1759.0 - throughput: 568.5048322910744 + inference_time: 1726.0 + throughput: 579.3742757821552 estimated_peak_memory_range: - min: 12288 - max: 2465832 + min: 16384 + max: 1310672 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 240 - job_id: jw56o88n5 + job_id: jw56032y5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1128.0 - throughput: 886.5248226950355 + inference_time: 1172.0 + throughput: 853.2423208191126 estimated_peak_memory_range: min: 634880 - max: 2124880 + max: 1948568 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jz5wyll4g + job_id: jz5wrdwmp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:45:13Z' + timestamp: '2024-08-27T00:02:09Z' - torchscript_onnx_tflite: - inference_time: 1760.0 - throughput: 568.1818181818181 + inference_time: 1726.0 + throughput: 579.3742757821552 estimated_peak_memory_range: - min: 12288 - max: 2485088 + min: 40960 + max: 1481840 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 240 - job_id: j1p3ozzmp + job_id: j1p3r4nnp job_status: Passed torchscript_onnx_qnn: - inference_time: 1138.0 - throughput: 878.7346221441124 + inference_time: 1164.0 + throughput: 859.106529209622 estimated_peak_memory_range: - min: 630784 - max: 1823472 + min: 638976 + max: 1922248 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: jmg9ozzmg + job_id: jmg9q308p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:45:14Z' + timestamp: '2024-08-27T00:02:10Z' - torchscript_onnx_qnn: - inference_time: 1320.0 - throughput: 757.5757575757576 + inference_time: 1345.0 + throughput: 743.4944237918215 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 186 - job_id: j7gj3rr1p + job_id: j7gj80de5 job_status: Passed torchscript_onnx: - inference_time: 2361.0 - throughput: 423.5493434985176 + inference_time: 2500.0 + throughput: 400.0 estimated_peak_memory_range: - min: 2519040 - max: 2519040 + min: 3026944 + max: 3026944 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 229 - job_id: jqp4eyy2g + job_id: jmg9q30mp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:45:18Z' + timestamp: '2024-08-27T00:02:14Z' diff --git a/qai_hub_models/models/squeezenet1_1/perf.yaml b/qai_hub_models/models/squeezenet1_1/perf.yaml index 80186941..a9dcbf97 100644 --- a/qai_hub_models/models/squeezenet1_1/perf.yaml +++ b/qai_hub_models/models/squeezenet1_1/perf.yaml @@ -45,11 +45,11 @@ models: - name: SqueezeNet-1_1 performance_metrics: - torchscript_onnx_tflite: - inference_time: 662.0 - throughput: 1510.5740181268882 + inference_time: 640.0 + throughput: 1562.5 estimated_peak_memory_range: - min: 12288 - max: 55504616 + min: 24576 + max: 1432472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jogk624v5 + job_id: jogkklxng job_status: Passed torchscript_onnx_qnn: - inference_time: 730.0 - throughput: 1369.86301369863 + inference_time: 709.0 + throughput: 1410.4372355430182 estimated_peak_memory_range: min: 622592 - max: 7050200 + max: 3712424 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: j7gj3r71p + job_id: j7gj80ve5 job_status: Passed torchscript_onnx: inference_time: 655.0 throughput: 1526.7175572519084 estimated_peak_memory_range: min: 12288 - max: 23775992 + max: 4130496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jqp4ey12g + job_id: jqp42xl1g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:44:36Z' + timestamp: '2024-08-27T00:01:36Z' - torchscript_onnx_tflite: - inference_time: 450.0 - throughput: 2222.222222222222 + inference_time: 441.0 + throughput: 2267.573696145125 estimated_peak_memory_range: min: 12288 - max: 25293616 + max: 25341200 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jn5q4lyeg + job_id: jn5qd7qog job_status: Passed torchscript_onnx_qnn: inference_time: 487.0 throughput: 2053.388090349076 estimated_peak_memory_range: - min: 0 - max: 11252112 + min: 618496 + max: 12046816 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jlpe67z8g + job_id: jlpenrdvp job_status: Passed torchscript_onnx: - inference_time: 480.0 - throughput: 2083.3333333333335 + inference_time: 483.0 + throughput: 2070.3933747412007 estimated_peak_memory_range: - min: 356352 - max: 27220768 + min: 0 + max: 26824720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: j0px0l48p + job_id: j0pxz7kl5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:44:37Z' + timestamp: '2024-08-27T00:01:37Z' - torchscript_onnx_tflite: - inference_time: 667.0 - throughput: 1499.2503748125937 + inference_time: 639.0 + throughput: 1564.9452269170579 estimated_peak_memory_range: min: 12288 - max: 5142144 + max: 1490808 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j1glwyx2p + job_id: j1glq0vmp job_status: Passed torchscript_onnx_qnn: - inference_time: 689.0 - throughput: 1451.3788098693758 + inference_time: 672.0 + throughput: 1488.095238095238 estimated_peak_memory_range: - min: 634880 - max: 2164096 + min: 655360 + max: 1769008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jz5wyl74g + job_id: jz5wrdemp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:44:30Z' + timestamp: '2024-08-27T00:01:32Z' - torchscript_onnx_tflite: - inference_time: 834.0 - throughput: 1199.0407673860911 + inference_time: 798.0 + throughput: 1253.1328320802006 estimated_peak_memory_range: - min: 16384 - max: 26955360 + min: 12288 + max: 27165440 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jw56o87n5 + job_id: jw5603yy5 job_status: Passed torchscript_onnx_qnn: - inference_time: 884.0 - throughput: 1131.2217194570135 + inference_time: 874.0 + throughput: 1144.1647597254005 estimated_peak_memory_range: min: 618496 - max: 15956480 + max: 15700032 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jz57oe4ng + job_id: jz578jy9p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:44:35Z' + timestamp: '2024-08-27T00:01:36Z' - torchscript_onnx_tflite: - inference_time: 661.0 - throughput: 1512.8593040847202 + inference_time: 640.0 + throughput: 1562.5 estimated_peak_memory_range: min: 12288 - max: 1432440 + max: 5770136 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j1p3oz9mp + job_id: j1p3r4jnp job_status: Passed torchscript_onnx_qnn: - inference_time: 687.0 - throughput: 1455.604075691412 + inference_time: 679.0 + throughput: 1472.7540500736377 estimated_peak_memory_range: - min: 647168 - max: 1750912 + min: 626688 + max: 1848880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jmg9ozmmg + job_id: jmg9q3l8p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:44:32Z' + timestamp: '2024-08-27T00:01:33Z' - torchscript_onnx_tflite: - inference_time: 672.0 - throughput: 1488.095238095238 + inference_time: 642.0 + throughput: 1557.632398753894 estimated_peak_memory_range: min: 24576 - max: 7231296 + max: 16907688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: jwgodlr15 + job_id: jwgo912kg job_status: Passed torchscript_onnx_qnn: - inference_time: 688.0 - throughput: 1453.4883720930231 + inference_time: 672.0 + throughput: 1488.095238095238 estimated_peak_memory_range: - min: 630784 - max: 1887176 + min: 28672 + max: 1408288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jnp1onjn5 + job_id: jnp1md47p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:44:33Z' + timestamp: '2024-08-27T00:01:34Z' - torchscript_onnx_tflite: - inference_time: 667.0 - throughput: 1499.2503748125937 + inference_time: 640.0 + throughput: 1562.5 estimated_peak_memory_range: - min: 36864 - max: 1442752 + min: 24576 + max: 1488264 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 41 - job_id: j1pv2ldzg + job_id: j1pvn16rg job_status: Passed torchscript_onnx_qnn: - inference_time: 682.0 - throughput: 1466.275659824047 + inference_time: 669.0 + throughput: 1494.7683109118086 estimated_peak_memory_range: - min: 36864 - max: 1537736 + min: 626688 + max: 2182008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jvgd6d36p + job_id: jvgdmrxzg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:44:34Z' + timestamp: '2024-08-27T00:01:35Z' - torchscript_onnx_qnn: - inference_time: 807.0 - throughput: 1239.1573729863692 + inference_time: 844.0 + throughput: 1184.8341232227488 estimated_peak_memory_range: - min: 602112 - max: 602112 + min: 1290240 + max: 1290240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 70 - job_id: jygzzlm4g + job_id: jygz0x3x5 job_status: Passed torchscript_onnx: - inference_time: 667.0 - throughput: 1499.2503748125937 + inference_time: 694.0 + throughput: 1440.922190201729 estimated_peak_memory_range: - min: 3751936 - max: 3751936 + min: 4116480 + max: 4116480 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 71 - job_id: jo5m90m7g + job_id: jo5mlwn9g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:44:38Z' + timestamp: '2024-08-27T00:01:38Z' diff --git a/qai_hub_models/models/squeezenet1_1_quantized/perf.yaml b/qai_hub_models/models/squeezenet1_1_quantized/perf.yaml index f2262a8c..814b7ff5 100644 --- a/qai_hub_models/models/squeezenet1_1_quantized/perf.yaml +++ b/qai_hub_models/models/squeezenet1_1_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: SqueezeNet-1_1Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 213.0 - throughput: 4694.835680751174 + inference_time: 202.0 + throughput: 4950.495049504951 estimated_peak_memory_range: - min: 12288 - max: 72497632 + min: 16384 + max: 1375160 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: j2p0om60p + job_id: j1p8k7vop job_status: Passed torchscript_onnx_qnn: - inference_time: 471.0 - throughput: 2123.1422505307855 + inference_time: 464.0 + throughput: 2155.1724137931033 estimated_peak_memory_range: - min: 20480 - max: 3081944 + min: 24576 + max: 10095272 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: j7gj3rq1p + job_id: jlpenrmvp job_status: Passed torchscript_onnx: - inference_time: 455.0 - throughput: 2197.802197802198 + inference_time: 462.0 + throughput: 2164.5021645021643 estimated_peak_memory_range: - min: 12288 - max: 2810304 + min: 36864 + max: 2980752 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 47 - job_id: j0px0lx8p + job_id: jo5mlwx9g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:43:58Z' + timestamp: '2024-08-27T00:01:03Z' - torchscript_onnx_tflite: - inference_time: 158.0 - throughput: 6329.113924050633 + inference_time: 150.0 + throughput: 6666.666666666667 estimated_peak_memory_range: - min: 0 - max: 25762016 + min: 12288 + max: 26084480 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: j1p8je1q5 + job_id: jogkkymng job_status: Passed torchscript_onnx_qnn: inference_time: 343.0 throughput: 2915.451895043732 estimated_peak_memory_range: min: 163840 - max: 13179984 + max: 12434224 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: jlpe67y8g + job_id: jygz0xdx5 job_status: Passed torchscript_onnx: inference_time: 362.0 throughput: 2762.4309392265195 estimated_peak_memory_range: - min: 16384 - max: 28531968 + min: 12288 + max: 28761488 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 47 - job_id: jo5m9087g + job_id: jegnw9vqg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:43:59Z' + timestamp: '2024-08-27T00:01:04Z' - torchscript_onnx_tflite: - inference_time: 209.0 - throughput: 4784.688995215311 + inference_time: 206.0 + throughput: 4854.368932038835 estimated_peak_memory_range: min: 12288 - max: 1317216 + max: 1196496 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: jogk628v5 + job_id: jn5qd2oog job_status: Passed torchscript_onnx_qnn: - inference_time: 426.0 - throughput: 2347.417840375587 + inference_time: 438.0 + throughput: 2283.10502283105 estimated_peak_memory_range: min: 184320 - max: 1716008 + max: 1629280 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: jz5wyl44g + job_id: jmg9q3n8p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:43:51Z' + timestamp: '2024-08-27T00:00:57Z' - torchscript_onnx_tflite: - inference_time: 254.0 - throughput: 3937.0078740157483 + inference_time: 240.0 + throughput: 4166.666666666667 estimated_peak_memory_range: - min: 12288 - max: 26640816 + min: 16384 + max: 26959360 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: jn5q4lveg + job_id: j1glq0mmp job_status: Passed torchscript_onnx_qnn: - inference_time: 523.0 - throughput: 1912.0458891013384 + inference_time: 528.0 + throughput: 1893.939393939394 estimated_peak_memory_range: min: 159744 - max: 13959824 + max: 14543936 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: jz57oe9ng + job_id: jqp42xr1g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:43:56Z' + timestamp: '2024-08-27T00:01:01Z' - torchscript_onnx_tflite: - inference_time: 231.0 - throughput: 4329.004329004329 + inference_time: 200.0 + throughput: 5000.0 estimated_peak_memory_range: - min: 16384 - max: 1458776 + min: 12288 + max: 9010904 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: j1glwyl2p + job_id: jw56034y5 job_status: Passed torchscript_onnx_qnn: - inference_time: 421.0 - throughput: 2375.296912114014 + inference_time: 424.0 + throughput: 2358.490566037736 estimated_peak_memory_range: min: 184320 - max: 1689352 + max: 1889328 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: jmg9ozdmg + job_id: jnp1mdz7p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:43:52Z' + timestamp: '2024-08-27T00:00:58Z' - torchscript_onnx_tflite: - inference_time: 208.0 - throughput: 4807.692307692308 + inference_time: 201.0 + throughput: 4975.124378109453 estimated_peak_memory_range: min: 12288 - max: 1295072 + max: 17489584 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: jw56o8wn5 + job_id: j1p3r40np job_status: Passed torchscript_onnx_qnn: inference_time: 428.0 throughput: 2336.448598130841 estimated_peak_memory_range: - min: 208896 - max: 1512496 + min: 176128 + max: 1315320 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: jnp1on6n5 + job_id: jvgdmr1zg job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:43:53Z' + timestamp: '2024-08-27T00:00:59Z' - torchscript_onnx_tflite: - inference_time: 214.0 - throughput: 4672.897196261682 + inference_time: 207.0 + throughput: 4830.917874396136 estimated_peak_memory_range: min: 12288 - max: 12864288 + max: 1883424 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: j1p3oz6mp + job_id: jwgo916kg job_status: Passed torchscript_onnx_qnn: - inference_time: 427.0 - throughput: 2341.92037470726 + inference_time: 425.0 + throughput: 2352.9411764705883 estimated_peak_memory_range: - min: 180224 - max: 1568328 + min: 176128 + max: 1680112 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: jvgd6d26p + job_id: jz578jr9p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:43:55Z' + timestamp: '2024-08-27T00:01:00Z' - torchscript_onnx_tflite: - inference_time: 527.0 - throughput: 1897.5332068311195 + inference_time: 498.0 + throughput: 2008.0321285140562 estimated_peak_memory_range: min: 16384 - max: 17354784 + max: 16795040 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: jwgodl815 + job_id: j1pvn1krg job_status: Passed torchscript_onnx_qnn: - inference_time: 990.0 - throughput: 1010.10101010101 + inference_time: 989.0 + throughput: 1011.1223458038422 estimated_peak_memory_range: min: 12288 - max: 8236656 + max: 8188032 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: jqp4ey32g + job_id: j0pxz7ol5 job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:43:57Z' + timestamp: '2024-08-27T00:01:02Z' - torchscript_onnx_tflite: - inference_time: 4107.0 - throughput: 243.48672997321646 + inference_time: 4166.0 + throughput: 240.03840614498318 estimated_peak_memory_range: - min: 90112 - max: 6812064 + min: 45056 + max: 7075456 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 43 - job_id: j1pv2l7zg + job_id: j7gj80ne5 job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:43:46Z' + timestamp: '2024-08-27T00:00:53Z' - torchscript_onnx_qnn: - inference_time: 534.0 - throughput: 1872.6591760299625 + inference_time: 557.0 + throughput: 1795.3321364452424 estimated_peak_memory_range: - min: 630784 - max: 630784 + min: 614400 + max: 614400 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 45 - job_id: jygzzln4g + job_id: jz5wrd6mp job_status: Passed torchscript_onnx: - inference_time: 512.0 - throughput: 1953.125 + inference_time: 501.0 + throughput: 1996.007984031936 estimated_peak_memory_range: - min: 4329472 - max: 4329472 + min: 2588672 + max: 2588672 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 47 - job_id: jegn1zkjp + job_id: jopr7437g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:44:00Z' + timestamp: '2024-08-27T00:01:04Z' diff --git a/qai_hub_models/models/swin_base/perf.yaml b/qai_hub_models/models/swin_base/perf.yaml index ad56001d..e7ece98a 100644 --- a/qai_hub_models/models/swin_base/perf.yaml +++ b/qai_hub_models/models/swin_base/perf.yaml @@ -45,26 +45,26 @@ models: - name: Swin-Base performance_metrics: - torchscript_onnx_tflite: - inference_time: 96955.0 - throughput: 10.31406322520757 + inference_time: 28173.0 + throughput: 35.49497746068931 estimated_peak_memory_range: - min: 9678848 - max: 19145656 + min: 0 + max: 2456664 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1511 - layers_on_gpu: 12 - layers_on_cpu: 45 + layers_on_npu: 1568 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1568 - job_id: jw56o8dy5 + job_id: jlpenll7p job_status: Passed torchscript_onnx_qnn: - inference_time: 31275.0 - throughput: 31.974420463629098 + inference_time: 31686.0 + throughput: 31.559679353657767 estimated_peak_memory_range: - min: 57344 - max: 51343152 + min: 110592 + max: 44406448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,22 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1255 - job_id: jz5wyl1mg - job_status: Passed - torchscript_onnx: - inference_time: 63799.0 - throughput: 15.674226868759698 - estimated_peak_memory_range: - min: 81920 - max: 237491528 - primary_compute_unit: NPU - precision: fp16 - layer_info: - layers_on_npu: 1141 - layers_on_gpu: 0 - layers_on_cpu: 0 - total_layers: 1141 - job_id: jz57oe7ng + job_id: jqp4299qg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,28 +81,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:42:55Z' + timestamp: '2024-08-27T11:48:17Z' - torchscript_onnx_tflite: - inference_time: 86552.0 - throughput: 11.553748035862833 + inference_time: 19636.0 + throughput: 50.926869016092894 estimated_peak_memory_range: - min: 9818112 - max: 470755216 + min: 36864 + max: 541281184 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1511 - layers_on_gpu: 12 - layers_on_cpu: 45 + layers_on_npu: 1568 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1568 - job_id: j1p3ozwnp + job_id: jygz044z5 job_status: Passed torchscript_onnx_qnn: - inference_time: 22004.0 - throughput: 45.44628249409198 + inference_time: 22100.0 + throughput: 45.248868778280546 estimated_peak_memory_range: min: 0 - max: 170014736 + max: 170028048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1255 - job_id: jmg9ozx8g + job_id: j0pxzddj5 job_status: Passed torchscript_onnx: - inference_time: 44087.0 - throughput: 22.682423390115 + inference_time: 44437.0 + throughput: 22.50376938137138 estimated_peak_memory_range: min: 0 - max: 786681472 + max: 783188160 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +125,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1141 - job_id: jqp4ey92g + job_id: jogkk98yg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,28 +134,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:42:56Z' + timestamp: '2024-08-27T11:48:25Z' - torchscript_onnx_tflite: - inference_time: 105207.0 - throughput: 9.505070955354682 + inference_time: 28063.0 + throughput: 35.63410896910523 estimated_peak_memory_range: - min: 9732096 - max: 14063208 + min: 69632 + max: 5746624 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1511 - layers_on_gpu: 12 - layers_on_cpu: 45 + layers_on_npu: 1568 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1568 - job_id: jwgodl4k5 + job_id: jz5wr11zp job_status: Passed torchscript_onnx_qnn: - inference_time: 28758.0 - throughput: 34.77293274914806 + inference_time: 29080.0 + throughput: 34.3878954607978 estimated_peak_memory_range: - min: 733184 - max: 2484936 + min: 724992 + max: 2527384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +163,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1255 - job_id: jvgd6dzzp + job_id: jegnw7kvg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,28 +172,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:42:49Z' + timestamp: '2024-08-27T11:48:20Z' - torchscript_onnx_tflite: - inference_time: 131305.0 - throughput: 7.615856212634705 + inference_time: 35229.0 + throughput: 28.385704958982657 estimated_peak_memory_range: - min: 9863168 - max: 445436864 + min: 114688 + max: 516633856 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1543 - layers_on_gpu: 2 - layers_on_cpu: 23 + layers_on_npu: 1568 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1568 - job_id: j1pv2l9rg + job_id: jmg9qxxqp job_status: Passed torchscript_onnx_qnn: - inference_time: 38020.0 - throughput: 26.301946344029457 + inference_time: 38198.0 + throughput: 26.179381119430335 estimated_peak_memory_range: - min: 679936 - max: 163251024 + min: 647168 + max: 164133600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +201,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1255 - job_id: jvgd6dz6p + job_id: j2p0xv62p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,28 +210,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:42:54Z' + timestamp: '2024-08-27T11:48:23Z' - torchscript_onnx_tflite: - inference_time: 101383.0 - throughput: 9.863586597358532 + inference_time: 28145.0 + throughput: 35.53028957186001 estimated_peak_memory_range: - min: 9793536 - max: 18674256 + min: 233472 + max: 4170928 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1511 - layers_on_gpu: 12 - layers_on_cpu: 45 + layers_on_npu: 1568 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1568 - job_id: j7gj3rwep + job_id: jnp1mvvkp job_status: Passed torchscript_onnx_qnn: - inference_time: 29293.0 - throughput: 34.13784863277916 + inference_time: 29192.0 + throughput: 34.25596053713346 estimated_peak_memory_range: - min: 663552 - max: 2221664 + min: 651264 + max: 1977256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +239,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1255 - job_id: jz5wyl14g + job_id: jopr7nwvg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,28 +248,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:42:51Z' + timestamp: '2024-08-27T11:48:21Z' - torchscript_onnx_tflite: - inference_time: 97303.0 - throughput: 10.277175421107263 + inference_time: 28195.0 + throughput: 35.46728143287817 estimated_peak_memory_range: - min: 9699328 - max: 19607080 + min: 49152 + max: 3510600 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1511 - layers_on_gpu: 12 - layers_on_cpu: 45 + layers_on_npu: 1568 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1568 - job_id: jlpe67lvg + job_id: jvgdmzzkg job_status: Passed torchscript_onnx_qnn: - inference_time: 29321.0 - throughput: 34.10524879778998 + inference_time: 29143.0 + throughput: 34.31355728648389 estimated_peak_memory_range: - min: 729088 - max: 2400912 + min: 704512 + max: 2000984 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +277,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1255 - job_id: jmg9ozxmg + job_id: jep2zvexp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,28 +286,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:42:52Z' + timestamp: '2024-08-27T11:48:22Z' - torchscript_onnx_tflite: - inference_time: 100429.0 - throughput: 9.957283254836751 + inference_time: 28332.0 + throughput: 35.29577862487646 estimated_peak_memory_range: - min: 9826304 - max: 13281992 + min: 57344 + max: 3609240 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1511 - layers_on_gpu: 12 - layers_on_cpu: 45 + layers_on_npu: 1568 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1568 - job_id: jygzzl4xg + job_id: jz57877qp job_status: Passed torchscript_onnx_qnn: - inference_time: 30201.0 - throughput: 33.111486374623354 + inference_time: 29075.0 + throughput: 34.393809114359414 estimated_peak_memory_range: - min: 667648 - max: 2260448 + min: 4550656 + max: 6161720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +315,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1255 - job_id: jnp1onvn5 + job_id: jqpyy7mrp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +324,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:42:53Z' + timestamp: '2024-08-27T11:48:23Z' - torchscript_onnx_qnn: - inference_time: 29676.0 - throughput: 33.69726378218089 + inference_time: 30498.0 + throughput: 32.78903534658011 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +338,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1255 - job_id: jnp1onv75 + job_id: jo5mlddyg job_status: Passed torchscript_onnx: - inference_time: 65625.0 - throughput: 15.238095238095237 + inference_time: 66472.0 + throughput: 15.043928270550007 estimated_peak_memory_range: - min: 208228352 - max: 208228352 + min: 208064512 + max: 208064512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +353,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1141 - job_id: j0px0ld8p + job_id: jn5qdmv7g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +362,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:42:57Z' + timestamp: '2024-08-27T11:48:26Z' diff --git a/qai_hub_models/models/swin_small/perf.yaml b/qai_hub_models/models/swin_small/perf.yaml index 110176cc..53a138b0 100644 --- a/qai_hub_models/models/swin_small/perf.yaml +++ b/qai_hub_models/models/swin_small/perf.yaml @@ -45,26 +45,26 @@ models: - name: Swin-Small performance_metrics: - torchscript_onnx_tflite: - inference_time: 64033.0 - throughput: 15.616947511439413 + inference_time: 21051.0 + throughput: 47.50368153531899 estimated_peak_memory_range: - min: 7258112 - max: 15100720 + min: 45056 + max: 3400104 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1508 - layers_on_gpu: 12 - layers_on_cpu: 43 + layers_on_npu: 1563 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1563 - job_id: jwgodlok5 + job_id: j1pvnwokg job_status: Passed torchscript_onnx_qnn: - inference_time: 23891.0 - throughput: 41.856766146247544 + inference_time: 23818.0 + throughput: 41.985053321017716 estimated_peak_memory_range: - min: 49152 - max: 44559608 + min: 12288 + max: 41413920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1246 - job_id: jnp1on775 + job_id: jvgdmv9rg job_status: Passed torchscript_onnx: - inference_time: 54362.0 - throughput: 18.39520253117987 + inference_time: 54734.0 + throughput: 18.270179413161838 estimated_peak_memory_range: - min: 159744 - max: 135883264 + min: 745472 + max: 1413882352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1136 - job_id: jep2or4qg + job_id: jo5mlz29g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,28 +96,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:42:05Z' + timestamp: '2024-08-26T23:59:15Z' - torchscript_onnx_tflite: - inference_time: 55328.0 - throughput: 18.074031231925968 + inference_time: 14545.0 + throughput: 68.75214850464077 estimated_peak_memory_range: - min: 7331840 - max: 376737632 + min: 45056 + max: 510912320 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1508 - layers_on_gpu: 12 - layers_on_cpu: 43 + layers_on_npu: 1563 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1563 - job_id: j1pv2lerg + job_id: j7gj8lmv5 job_status: Passed torchscript_onnx_qnn: - inference_time: 16213.0 - throughput: 61.678899648430274 + inference_time: 16315.0 + throughput: 61.29328838492185 estimated_peak_memory_range: - min: 622592 - max: 141395664 + min: 0 + max: 136814080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1246 - job_id: jvgd6d8zp + job_id: jz5wr9vmp job_status: Passed torchscript_onnx: - inference_time: 37876.0 - throughput: 26.401943183018272 + inference_time: 38103.0 + throughput: 26.24465265202215 estimated_peak_memory_range: - min: 0 - max: 740003632 + min: 753664 + max: 741642176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1136 - job_id: jqpy8oqlg + job_id: jegnweyqg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,28 +149,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:42:06Z' + timestamp: '2024-08-26T23:59:16Z' - torchscript_onnx_tflite: - inference_time: 64486.0 - throughput: 15.507241881958874 + inference_time: 20941.0 + throughput: 47.753211403466885 estimated_peak_memory_range: - min: 16384 - max: 23046768 + min: 45056 + max: 2745928 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1508 - layers_on_gpu: 12 - layers_on_cpu: 43 + layers_on_npu: 1563 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1563 - job_id: j7gj3roep + job_id: jlpenv1op job_status: Passed torchscript_onnx_qnn: - inference_time: 21421.0 - throughput: 46.683161383688905 + inference_time: 21574.0 + throughput: 46.35209047928061 estimated_peak_memory_range: min: 684032 - max: 1862232 + max: 2209056 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1246 - job_id: jqp4eym1g + job_id: jnp1m8l7p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,28 +187,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:41:59Z' + timestamp: '2024-08-26T23:59:11Z' - torchscript_onnx_tflite: - inference_time: 83937.0 - throughput: 11.913697177645139 + inference_time: 26502.0 + throughput: 37.73300128292205 estimated_peak_memory_range: - min: 7274496 - max: 359619840 + min: 65536 + max: 494493760 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1538 - layers_on_gpu: 2 - layers_on_cpu: 23 + layers_on_npu: 1563 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1563 - job_id: jlpe678vg + job_id: jygz079o5 job_status: Passed torchscript_onnx_qnn: - inference_time: 28706.0 - throughput: 34.83592280359507 + inference_time: 28755.0 + throughput: 34.77656059815684 estimated_peak_memory_range: - min: 651264 - max: 134165840 + min: 0 + max: 135708592 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1246 - job_id: joprxlo7p + job_id: j0pxz1jl5 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,28 +225,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:42:04Z' + timestamp: '2024-08-26T23:59:14Z' - torchscript_onnx_tflite: - inference_time: 64341.0 - throughput: 15.542189272780965 + inference_time: 21067.0 + throughput: 47.46760336070632 estimated_peak_memory_range: - min: 7266304 - max: 13357208 + min: 32768 + max: 3424600 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1508 - layers_on_gpu: 12 - layers_on_cpu: 43 + layers_on_npu: 1563 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1563 - job_id: jygzzl8xg + job_id: jz5wr9v3p job_status: Passed torchscript_onnx_qnn: - inference_time: 21865.0 - throughput: 45.7351932311914 + inference_time: 22008.0 + throughput: 45.43802253725918 estimated_peak_memory_range: - min: 4673536 - max: 6287056 + min: 675840 + max: 1986240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1246 - job_id: j0px0l3lp + job_id: jvgdmv9zg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,28 +263,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:42:00Z' + timestamp: '2024-08-26T23:59:12Z' - torchscript_onnx_tflite: - inference_time: 63876.0 - throughput: 15.655332206149415 + inference_time: 21026.0 + throughput: 47.56016360696281 estimated_peak_memory_range: - min: 7270400 - max: 15315808 + min: 36864 + max: 3334344 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1508 - layers_on_gpu: 12 - layers_on_cpu: 43 + layers_on_npu: 1563 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1563 - job_id: jz5wyl8mg + job_id: jmg9q41wp job_status: Passed torchscript_onnx_qnn: - inference_time: 21751.0 - throughput: 45.974897705852605 + inference_time: 21998.0 + throughput: 45.45867806164197 estimated_peak_memory_range: - min: 700416 - max: 1867376 + min: 684032 + max: 1970400 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1246 - job_id: jo5m90o9g + job_id: jz578dw9p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,28 +301,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:42:01Z' + timestamp: '2024-08-26T23:59:13Z' - torchscript_onnx_tflite: - inference_time: 66495.0 - throughput: 15.03872471614407 + inference_time: 20985.0 + throughput: 47.65308553728854 estimated_peak_memory_range: - min: 290816 - max: 8286632 + min: 40960 + max: 2801224 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1508 - layers_on_gpu: 12 - layers_on_cpu: 43 + layers_on_npu: 1563 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1563 - job_id: jmg9ozk8g + job_id: jnp1m8l8p job_status: Passed torchscript_onnx_qnn: - inference_time: 23010.0 - throughput: 43.459365493263796 + inference_time: 22064.0 + throughput: 45.32269760696157 estimated_peak_memory_range: - min: 692224 - max: 2000064 + min: 724992 + max: 2091456 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1246 - job_id: jegn1zoqp + job_id: jqp42wo1g job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:42:03Z' + timestamp: '2024-08-26T23:59:13Z' - torchscript_onnx_qnn: - inference_time: 22248.0 - throughput: 44.94786048184106 + inference_time: 22532.0 + throughput: 44.381324338718265 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,11 +353,11 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1246 - job_id: jz57oek9g + job_id: jmg9q418p job_status: Passed torchscript_onnx: - inference_time: 56643.0 - throughput: 17.65443214519005 + inference_time: 56984.0 + throughput: 17.548785624034817 estimated_peak_memory_range: min: 124555264 max: 124555264 @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1136 - job_id: j2p0omdnp + job_id: jopr7yq7g job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:42:07Z' + timestamp: '2024-08-26T23:59:17Z' diff --git a/qai_hub_models/models/swin_tiny/perf.yaml b/qai_hub_models/models/swin_tiny/perf.yaml index 4d501e10..6dbf8690 100644 --- a/qai_hub_models/models/swin_tiny/perf.yaml +++ b/qai_hub_models/models/swin_tiny/perf.yaml @@ -45,26 +45,26 @@ models: - name: Swin-Tiny performance_metrics: - torchscript_onnx_tflite: - inference_time: 37166.0 - throughput: 26.90631222084701 + inference_time: 13486.0 + throughput: 74.15097137772504 estimated_peak_memory_range: - min: 7282688 - max: 15272000 + min: 20480 + max: 2689624 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 794 - layers_on_gpu: 12 - layers_on_cpu: 31 + layers_on_npu: 837 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 837 - job_id: j7gj3r1ep + job_id: j7gj8lxv5 job_status: Passed torchscript_onnx_qnn: - inference_time: 14990.0 - throughput: 66.711140760507 + inference_time: 15104.0 + throughput: 66.20762711864407 estimated_peak_memory_range: - min: 16384 - max: 24112800 + min: 40960 + max: 25697016 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 700 - job_id: jz57oe69g + job_id: jz578dzvp job_status: Passed torchscript_onnx: - inference_time: 32363.0 - throughput: 30.899483978617557 + inference_time: 32562.0 + throughput: 30.71064430931761 estimated_peak_memory_range: - min: 69632 - max: 69443992 + min: 73728 + max: 69131504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 614 - job_id: j2p0omknp + job_id: j2p0xry9p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,28 +96,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:41:18Z' + timestamp: '2024-08-26T23:58:32Z' - torchscript_onnx_tflite: - inference_time: 31557.0 - throughput: 31.688690306429635 + inference_time: 9191.0 + throughput: 108.80208900010881 estimated_peak_memory_range: - min: 7299072 - max: 240854208 + min: 49152 + max: 318471024 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 794 - layers_on_gpu: 12 - layers_on_cpu: 31 + layers_on_npu: 837 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 837 - job_id: jlpe672vg + job_id: jlpenv9op job_status: Passed torchscript_onnx_qnn: - inference_time: 10045.0 - throughput: 99.55201592832255 + inference_time: 10196.0 + throughput: 98.07767752059631 estimated_peak_memory_range: - min: 618496 - max: 90976496 + min: 143360 + max: 96806640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 700 - job_id: jqp4ey81g + job_id: jqp42wq8g job_status: Passed torchscript_onnx: - inference_time: 22517.0 - throughput: 44.41088955011769 + inference_time: 22541.0 + throughput: 44.36360409919702 estimated_peak_memory_range: min: 0 - max: 429819328 + max: 426693712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 614 - job_id: j1p8je8o5 + job_id: j1p8k7okp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,28 +149,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:41:19Z' + timestamp: '2024-08-26T23:58:33Z' - torchscript_onnx_tflite: - inference_time: 39662.0 - throughput: 25.213050274822248 + inference_time: 13400.0 + throughput: 74.6268656716418 estimated_peak_memory_range: - min: 7254016 - max: 13603288 + min: 24576 + max: 2559912 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 794 - layers_on_gpu: 12 - layers_on_cpu: 31 + layers_on_npu: 837 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 837 - job_id: jygzzlwxg + job_id: jygz07eo5 job_status: Passed torchscript_onnx_qnn: - inference_time: 13216.0 - throughput: 75.66585956416465 + inference_time: 13271.0 + throughput: 75.35227187099692 estimated_peak_memory_range: - min: 634880 - max: 1852472 + min: 643072 + max: 1927696 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 700 - job_id: jo5m9049g + job_id: jo5mlzrdg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,28 +187,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:41:12Z' + timestamp: '2024-08-26T23:58:28Z' - torchscript_onnx_tflite: - inference_time: 47527.0 - throughput: 21.040671618238054 + inference_time: 16878.0 + throughput: 59.248726152387725 estimated_peak_memory_range: - min: 7421952 - max: 230274336 + min: 57344 + max: 308412224 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 824 - layers_on_gpu: 2 - layers_on_cpu: 11 + layers_on_npu: 837 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 837 - job_id: jz5wylxmg + job_id: jz5wr9o3p job_status: Passed torchscript_onnx_qnn: - inference_time: 18052.0 - throughput: 55.39552404165743 + inference_time: 18096.0 + throughput: 55.260831122900086 estimated_peak_memory_range: - min: 622592 - max: 92718288 + min: 634880 + max: 96200048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 700 - job_id: jqpy8onlg + job_id: jqpyyde8p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,28 +225,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:41:17Z' + timestamp: '2024-08-26T23:58:32Z' - torchscript_onnx_tflite: - inference_time: 39668.0 - throughput: 25.209236664313803 + inference_time: 13390.0 + throughput: 74.68259895444362 estimated_peak_memory_range: - min: 7254016 - max: 13612888 + min: 24576 + max: 2580104 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 794 - layers_on_gpu: 12 - layers_on_cpu: 31 + layers_on_npu: 837 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 837 - job_id: jmg9oz88g + job_id: jmg9q4vwp job_status: Passed torchscript_onnx_qnn: - inference_time: 13366.0 - throughput: 74.81669908723627 + inference_time: 13469.0 + throughput: 74.24456158586383 estimated_peak_memory_range: - min: 667648 - max: 1892624 + min: 704512 + max: 2062040 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 700 - job_id: jegn1zxqp + job_id: jegnwe2kg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,28 +263,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:41:13Z' + timestamp: '2024-08-26T23:58:29Z' - torchscript_onnx_tflite: - inference_time: 37323.0 - throughput: 26.793130241406104 + inference_time: 13401.0 + throughput: 74.62129691814043 estimated_peak_memory_range: - min: 7294976 - max: 14015240 + min: 45056 + max: 2624000 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 794 - layers_on_gpu: 12 - layers_on_cpu: 31 + layers_on_npu: 837 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 837 - job_id: jnp1on375 + job_id: jnp1m808p job_status: Passed torchscript_onnx_qnn: - inference_time: 13546.0 - throughput: 73.82253063635021 + inference_time: 13493.0 + throughput: 74.11250277921886 estimated_peak_memory_range: - min: 667648 - max: 1865880 + min: 684032 + max: 2109560 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 700 - job_id: joprxl97p + job_id: jopr7yk0g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,28 +301,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:41:14Z' + timestamp: '2024-08-26T23:58:30Z' - torchscript_onnx_tflite: - inference_time: 39187.0 - throughput: 25.51866690484089 + inference_time: 13356.0 + throughput: 74.87271638215034 estimated_peak_memory_range: - min: 4231168 - max: 12819912 + min: 24576 + max: 5485248 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 794 - layers_on_gpu: 12 - layers_on_cpu: 31 + layers_on_npu: 837 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 837 - job_id: jvgd6d0zp + job_id: jvgdmvwrg job_status: Passed torchscript_onnx_qnn: - inference_time: 13747.0 - throughput: 72.7431439586819 + inference_time: 13531.0 + throughput: 73.90436774813392 estimated_peak_memory_range: - min: 671744 - max: 1845672 + min: 675840 + max: 2338032 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 700 - job_id: jep2orjqg + job_id: jep2zm8rp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:41:15Z' + timestamp: '2024-08-26T23:58:31Z' - torchscript_onnx_qnn: - inference_time: 13820.0 - throughput: 72.3589001447178 + inference_time: 13868.0 + throughput: 72.10845111047014 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 700 - job_id: j0px0lmlp + job_id: j0pxz1v35 job_status: Passed torchscript_onnx: - inference_time: 33728.0 - throughput: 29.648956356736242 + inference_time: 33860.0 + throughput: 29.533372711163615 estimated_peak_memory_range: - min: 68239360 - max: 68239360 + min: 68112384 + max: 68112384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 614 - job_id: jogk62dn5 + job_id: jogkkyzwg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:41:20Z' + timestamp: '2024-08-26T23:58:34Z' diff --git a/qai_hub_models/models/trocr/export.py b/qai_hub_models/models/trocr/export.py index da9d40bc..d79f58a0 100644 --- a/qai_hub_models/models/trocr/export.py +++ b/qai_hub_models/models/trocr/export.py @@ -128,8 +128,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/trocr/perf.yaml b/qai_hub_models/models/trocr/perf.yaml index cb1b708c..2085d9c5 100644 --- a/qai_hub_models/models/trocr/perf.yaml +++ b/qai_hub_models/models/trocr/perf.yaml @@ -45,26 +45,26 @@ models: - name: TrOCREncoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 181040.0 - throughput: 5.52364118426867 + inference_time: 118610.0 + throughput: 8.430992327796982 estimated_peak_memory_range: - min: 8560640 - max: 16490968 + min: 7258112 + max: 9862032 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 579 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 591 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 591 - job_id: jmg9oz38g + job_id: jmg9q4rwp job_status: Passed torchscript_onnx_qnn: - inference_time: 120369.0 - throughput: 8.307786888650732 + inference_time: 120917.0 + throughput: 8.270135712927049 estimated_peak_memory_range: - min: 1806336 - max: 23161328 + min: 1900544 + max: 22214408 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 443 - job_id: jn5q4l7og + job_id: jn5qd2nng job_status: Passed torchscript_onnx: - inference_time: 108220.0 - throughput: 9.240436148586213 + inference_time: 108414.0 + throughput: 9.223900972199162 estimated_peak_memory_range: - min: 16171008 - max: 18811152 + min: 180224 + max: 118712344 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 396 - job_id: jo5m90v9g + job_id: jo5mlzjdg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,28 +96,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:40:25Z' + timestamp: '2024-08-26T23:57:44Z' - torchscript_onnx_tflite: - inference_time: 133140.0 - throughput: 7.510890791647889 + inference_time: 89287.0 + throughput: 11.199838722322399 estimated_peak_memory_range: - min: 9650176 - max: 307591040 + min: 5795840 + max: 324455872 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 579 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 591 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 591 - job_id: jvgd6drzp + job_id: jvgdmvkrg job_status: Passed torchscript_onnx_qnn: - inference_time: 90562.0 - throughput: 11.04215896292043 + inference_time: 90872.0 + throughput: 11.004489831851396 estimated_peak_memory_range: - min: 0 - max: 65159552 + min: 1785856 + max: 66784144 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 443 - job_id: jw56o8qy5 + job_id: jw5601x65 job_status: Passed torchscript_onnx: - inference_time: 82687.0 - throughput: 12.09379950899174 + inference_time: 81833.0 + throughput: 12.220009042806693 estimated_peak_memory_range: min: 0 - max: 356258544 + max: 352725232 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 396 - job_id: joprxl17p + job_id: jopr7yz0g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,28 +149,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:40:27Z' + timestamp: '2024-08-26T23:57:45Z' - torchscript_onnx_tflite: - inference_time: 170147.0 - throughput: 5.8772708305171415 + inference_time: 118068.0 + throughput: 8.469695429752345 estimated_peak_memory_range: - min: 10792960 - max: 18121968 + min: 11644928 + max: 773467880 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 579 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 591 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 591 - job_id: jqp4eyx1g + job_id: jqp42w78g job_status: Passed torchscript_onnx_qnn: - inference_time: 101060.0 - throughput: 9.895111814763506 + inference_time: 102469.0 + throughput: 9.759049078257815 estimated_peak_memory_range: - min: 1949696 - max: 8410160 + min: 1974272 + max: 8295040 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 443 - job_id: j7gj3rkep + job_id: j7gj8ljv5 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,28 +187,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:40:15Z' + timestamp: '2024-08-26T23:57:35Z' - torchscript_onnx_tflite: - inference_time: 208151.0 - throughput: 4.804204639900841 + inference_time: 125662.0 + throughput: 7.957855198866802 estimated_peak_memory_range: - min: 1961984 - max: 270124272 + min: 7376896 + max: 310806560 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 579 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 591 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 591 - job_id: jo5m90w9g + job_id: jo5mlz7dg job_status: Passed torchscript_onnx_qnn: - inference_time: 127551.0 - throughput: 7.8400012544002005 + inference_time: 125196.0 + throughput: 7.987475638199303 estimated_peak_memory_range: min: 0 - max: 63551104 + max: 60173376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 443 - job_id: jqp4eyj1g + job_id: jqp42wz8g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,28 +225,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:40:23Z' + timestamp: '2024-08-26T23:57:42Z' - torchscript_onnx_tflite: - inference_time: 170218.0 - throughput: 5.8748193493050085 + inference_time: 118401.0 + throughput: 8.445874612545502 estimated_peak_memory_range: - min: 8478720 - max: 16510864 + min: 7258112 + max: 9503704 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 579 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 591 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 591 - job_id: joprxl47p + job_id: jopr7yr0g job_status: Passed torchscript_onnx_qnn: - inference_time: 101898.0 - throughput: 9.813735303931383 + inference_time: 103970.0 + throughput: 9.618159084351255 estimated_peak_memory_range: - min: 2023424 - max: 3693104 + min: 1994752 + max: 3664168 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 443 - job_id: jygzzlvxg + job_id: jygz071o5 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,28 +263,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:40:17Z' + timestamp: '2024-08-26T23:57:37Z' - torchscript_onnx_tflite: - inference_time: 180094.0 - throughput: 5.552655835286017 + inference_time: 118463.0 + throughput: 8.441454293745727 estimated_peak_memory_range: - min: 10817536 - max: 18394416 + min: 7274496 + max: 9387912 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 579 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 591 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 591 - job_id: jqpy8o4lg + job_id: jqpyydl8p job_status: Passed torchscript_onnx_qnn: - inference_time: 101728.0 - throughput: 9.830135262661214 + inference_time: 103168.0 + throughput: 9.692928039702233 estimated_peak_memory_range: - min: 1912832 - max: 3876360 + min: 1994752 + max: 8729296 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 443 - job_id: jmg9oz98g + job_id: jmg9q46wp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,28 +301,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:40:19Z' + timestamp: '2024-08-26T23:57:39Z' - torchscript_onnx_tflite: - inference_time: 172118.0 - throughput: 5.809967580380902 + inference_time: 118708.0 + throughput: 8.424032078714156 estimated_peak_memory_range: - min: 6406144 - max: 13598912 + min: 7282688 + max: 9367064 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 579 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 591 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 591 - job_id: j1p8je3o5 + job_id: j1p8k7nkp job_status: Passed torchscript_onnx_qnn: - inference_time: 101063.0 - throughput: 9.89481808376953 + inference_time: 102837.0 + throughput: 9.724126530334413 estimated_peak_memory_range: - min: 1961984 - max: 8187024 + min: 1953792 + max: 10242888 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 443 - job_id: jvgd6d7zp + job_id: jvgdmvjrg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:40:21Z' + timestamp: '2024-08-26T23:57:41Z' - torchscript_onnx_qnn: - inference_time: 101677.0 - throughput: 9.835065944117156 + inference_time: 103205.0 + throughput: 9.689453030376436 estimated_peak_memory_range: min: 1773568 max: 1773568 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 443 - job_id: jwgodlek5 + job_id: jwgo9vxqg job_status: Passed torchscript_onnx: - inference_time: 109899.0 - throughput: 9.099263869552953 + inference_time: 110122.0 + throughput: 9.080837616461743 estimated_peak_memory_range: - min: 115023872 - max: 115023872 + min: 114962432 + max: 114962432 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 396 - job_id: jqpy8ovlg + job_id: jqpyyd98p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,53 +377,53 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:40:29Z' + timestamp: '2024-08-26T23:57:47Z' - name: TrOCRDecoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 6490.0 - throughput: 154.08320493066256 + inference_time: 2671.0 + throughput: 374.3916136278547 estimated_peak_memory_range: - min: 7139328 - max: 13687032 + min: 16384 + max: 2428296 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 375 - layers_on_gpu: 1 - layers_on_cpu: 6 - total_layers: 382 - job_id: jnp1ond75 + layers_on_npu: 399 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 399 + job_id: jnp1m898p job_status: Passed torchscript_onnx_qnn: - inference_time: 2958.0 - throughput: 338.0662609871535 + inference_time: 3212.0 + throughput: 311.332503113325 estimated_peak_memory_range: - min: 28672 - max: 125473456 + min: 614400 + max: 284943552 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 334 + layers_on_npu: 357 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 334 - job_id: j1glwyemp + total_layers: 357 + job_id: j1glqkdjp job_status: Passed torchscript_onnx: - inference_time: 3146.0 - throughput: 317.86395422759057 + inference_time: 3397.0 + throughput: 294.3773918163085 estimated_peak_memory_range: - min: 0 - max: 78236048 + min: 7503872 + max: 10025712 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 356 + layers_on_npu: 379 layers_on_gpu: 0 layers_on_cpu: 1 - total_layers: 357 - job_id: jegn1zrqp + total_layers: 380 + job_id: jegnwejkg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -432,51 +432,51 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:40:26Z' + timestamp: '2024-08-26T23:57:44Z' - torchscript_onnx_tflite: - inference_time: 5291.0 - throughput: 189.000189000189 + inference_time: 1916.0 + throughput: 521.9206680584551 estimated_peak_memory_range: - min: 5459968 - max: 86074112 + min: 12288 + max: 195705056 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 375 - layers_on_gpu: 1 - layers_on_cpu: 6 - total_layers: 382 - job_id: jz57oej9g + layers_on_npu: 399 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 399 + job_id: jz578dmvp job_status: Passed torchscript_onnx_qnn: - inference_time: 2165.0 - throughput: 461.8937644341801 + inference_time: 2241.0 + throughput: 446.2293618920125 estimated_peak_memory_range: min: 0 - max: 52541536 + max: 47854176 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 334 + layers_on_npu: 357 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 334 - job_id: j1p3ozqnp + total_layers: 357 + job_id: j1p3rmd3p job_status: Passed torchscript_onnx: - inference_time: 2362.0 - throughput: 423.3700254022015 + inference_time: 2709.0 + throughput: 369.139904023625 estimated_peak_memory_range: - min: 0 - max: 150697024 + min: 8478720 + max: 159710128 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 356 + layers_on_npu: 379 layers_on_gpu: 0 layers_on_cpu: 1 - total_layers: 357 - job_id: jep2or3qg + total_layers: 380 + job_id: jep2zm2rp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -485,36 +485,36 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:40:28Z' + timestamp: '2024-08-26T23:57:46Z' - torchscript_onnx_tflite: - inference_time: 6386.0 - throughput: 156.59254619480112 + inference_time: 2657.0 + throughput: 376.3643206624012 estimated_peak_memory_range: - min: 7135232 - max: 14014936 + min: 12288 + max: 2243152 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 375 - layers_on_gpu: 1 - layers_on_cpu: 6 - total_layers: 382 - job_id: j0px0l7lp + layers_on_npu: 399 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 399 + job_id: j0pxz1q35 job_status: Passed torchscript_onnx_qnn: - inference_time: 2529.0 - throughput: 395.41320680110715 + inference_time: 2685.0 + throughput: 372.43947858472995 estimated_peak_memory_range: - min: 1232896 - max: 2922680 + min: 737280 + max: 2865680 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 334 + layers_on_npu: 357 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 334 - job_id: jlpe674vg + total_layers: 357 + job_id: jlpenvjop job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -523,36 +523,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:40:15Z' + timestamp: '2024-08-26T23:57:35Z' - torchscript_onnx_tflite: - inference_time: 10409.0 - throughput: 96.07070804111827 + inference_time: 2814.0 + throughput: 355.36602700781805 estimated_peak_memory_range: - min: 7204864 - max: 85579360 + min: 12288 + max: 194789120 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 378 - layers_on_gpu: 1 - layers_on_cpu: 3 - total_layers: 382 - job_id: jegn1z9qp + layers_on_npu: 399 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 399 + job_id: jegnwe4kg job_status: Passed torchscript_onnx_qnn: - inference_time: 3309.0 - throughput: 302.2061045633122 + inference_time: 3445.0 + throughput: 290.2757619738752 estimated_peak_memory_range: - min: 5382144 - max: 52996512 + min: 5636096 + max: 51838096 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 334 + layers_on_npu: 357 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 334 - job_id: j0px0lelp + total_layers: 357 + job_id: j0pxz1w35 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -561,36 +561,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:40:24Z' + timestamp: '2024-08-26T23:57:43Z' - torchscript_onnx_tflite: - inference_time: 6422.0 - throughput: 155.71473061351602 + inference_time: 2648.0 + throughput: 377.64350453172204 estimated_peak_memory_range: - min: 7135232 - max: 13710368 + min: 12288 + max: 2048000 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 375 - layers_on_gpu: 1 - layers_on_cpu: 6 - total_layers: 382 - job_id: jep2or7qg + layers_on_npu: 399 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 399 + job_id: jep2zm1rp job_status: Passed torchscript_onnx_qnn: - inference_time: 2557.0 - throughput: 391.08330074305826 + inference_time: 2708.0 + throughput: 369.2762186115214 estimated_peak_memory_range: - min: 2424832 - max: 3714344 + min: 1921024 + max: 4005632 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 334 + layers_on_npu: 357 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 334 - job_id: jz5wylmmg + total_layers: 357 + job_id: jz5wr9j3p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -599,36 +599,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:40:17Z' + timestamp: '2024-08-26T23:57:37Z' - torchscript_onnx_tflite: - inference_time: 6464.0 - throughput: 154.7029702970297 + inference_time: 2704.0 + throughput: 369.8224852071006 estimated_peak_memory_range: - min: 7135232 - max: 12387968 + min: 12288 + max: 2421432 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 375 - layers_on_gpu: 1 - layers_on_cpu: 6 - total_layers: 382 - job_id: j2p0om1np + layers_on_npu: 399 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 399 + job_id: j2p0xrw9p job_status: Passed torchscript_onnx_qnn: - inference_time: 2541.0 - throughput: 393.5458480913026 + inference_time: 3004.0 + throughput: 332.88948069241013 estimated_peak_memory_range: - min: 1863680 - max: 3210880 + min: 1310720 + max: 2620152 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 334 + layers_on_npu: 357 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 334 - job_id: jnp1onq75 + total_layers: 357 + job_id: jnp1m8r8p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -637,36 +637,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:40:19Z' + timestamp: '2024-08-26T23:57:39Z' - torchscript_onnx_tflite: - inference_time: 6390.0 - throughput: 156.49452269170578 + inference_time: 2674.0 + throughput: 373.97157816005983 estimated_peak_memory_range: - min: 7135232 - max: 13284968 + min: 12288 + max: 1966768 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 375 - layers_on_gpu: 1 - layers_on_cpu: 6 - total_layers: 382 - job_id: jogk62ln5 + layers_on_npu: 399 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 399 + job_id: jogkky1wg job_status: Passed torchscript_onnx_qnn: - inference_time: 2572.0 - throughput: 388.8024883359254 + inference_time: 2721.0 + throughput: 367.5119441381845 estimated_peak_memory_range: - min: 1413120 - max: 3143072 + min: 1269760 + max: 2578240 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 334 + layers_on_npu: 357 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 334 - job_id: jz57oev9g + total_layers: 357 + job_id: jz578dqvp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -675,36 +675,36 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:40:22Z' + timestamp: '2024-08-26T23:57:41Z' - torchscript_onnx_qnn: - inference_time: 2792.0 - throughput: 358.1661891117478 + inference_time: 2956.0 + throughput: 338.29499323410016 estimated_peak_memory_range: - min: 7176192 - max: 7176192 + min: 7393280 + max: 7393280 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 334 + layers_on_npu: 357 layers_on_gpu: 0 layers_on_cpu: 0 - total_layers: 334 - job_id: j1pv2lzrg + total_layers: 357 + job_id: j1pvnwjkg job_status: Passed torchscript_onnx: - inference_time: 3026.0 - throughput: 330.4692663582287 + inference_time: 3187.0 + throughput: 313.7747097583935 estimated_peak_memory_range: - min: 73543680 - max: 73543680 + min: 72073216 + max: 72073216 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 356 + layers_on_npu: 379 layers_on_gpu: 0 layers_on_cpu: 1 - total_layers: 357 - job_id: j2p0omenp + total_layers: 380 + job_id: j2p0xrn9p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -713,4 +713,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:40:30Z' + timestamp: '2024-08-26T23:57:48Z' diff --git a/qai_hub_models/models/unet_segmentation/perf.yaml b/qai_hub_models/models/unet_segmentation/perf.yaml index 7162c87b..90e49486 100644 --- a/qai_hub_models/models/unet_segmentation/perf.yaml +++ b/qai_hub_models/models/unet_segmentation/perf.yaml @@ -45,11 +45,11 @@ models: - name: Unet-Segmentation performance_metrics: - torchscript_onnx_tflite: - inference_time: 154294.0 - throughput: 6.481133420612597 + inference_time: 158663.0 + throughput: 6.302666658263111 estimated_peak_memory_range: - min: 6172672 - max: 8551904 + min: 6623232 + max: 110677496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 32 - job_id: jo5m90zdg + job_id: jvgdmvyrg job_status: Passed torchscript_onnx_qnn: - inference_time: 151839.0 - throughput: 6.585923247650472 + inference_time: 156066.0 + throughput: 6.407545525610959 estimated_peak_memory_range: - min: 9981952 - max: 30876984 + min: 9957376 + max: 27485240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 52 - job_id: jogk62yw5 + job_id: jep2zmqrp job_status: Passed torchscript_onnx: - inference_time: 159287.0 - throughput: 6.277976231581987 + inference_time: 156502.0 + throughput: 6.389694700387215 estimated_peak_memory_range: - min: 45056 - max: 1595227760 + min: 20480 + max: 58473536 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jlpe67rog + job_id: j1p3rml3p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:39:18Z' + timestamp: '2024-08-26T23:56:42Z' - torchscript_onnx_tflite: - inference_time: 113749.0 - throughput: 8.79128607724024 + inference_time: 112095.0 + throughput: 8.921004505107275 estimated_peak_memory_range: - min: 5398528 - max: 344167152 + min: 5386240 + max: 344870544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 32 - job_id: jegn1zekp + job_id: jz578d1vp job_status: Passed torchscript_onnx_qnn: - inference_time: 110426.0 - throughput: 9.055838298951334 + inference_time: 110551.0 + throughput: 9.045598863872783 estimated_peak_memory_range: - min: 9867264 - max: 85222912 + min: 9961472 + max: 80448288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 52 - job_id: jn5q4l2ng + job_id: jqpyydk8p job_status: Passed torchscript_onnx: - inference_time: 118949.0 - throughput: 8.406964329250352 + inference_time: 118194.0 + throughput: 8.460666362082678 estimated_peak_memory_range: - min: 24854528 - max: 374193264 + min: 884736 + max: 348008176 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jygzzlxog + job_id: jwgo9v7qg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:39:19Z' + timestamp: '2024-08-26T23:56:43Z' - torchscript_onnx_tflite: - inference_time: 158865.0 - throughput: 6.2946526925376896 + inference_time: 154304.0 + throughput: 6.480713396930734 estimated_peak_memory_range: - min: 6705152 - max: 463295672 + min: 6713344 + max: 228465840 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 32 - job_id: joprxly0p + job_id: jqp42w68g job_status: Passed torchscript_onnx_qnn: - inference_time: 137288.0 - throughput: 7.283957811316357 + inference_time: 141752.0 + throughput: 7.054574185902139 estimated_peak_memory_range: - min: 10121216 - max: 11648752 + min: 9883648 + max: 11067008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 52 - job_id: jw56o8165 + job_id: j1p8k7dkp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:39:12Z' + timestamp: '2024-08-26T23:56:38Z' - torchscript_onnx_tflite: - inference_time: 280767.0 - throughput: 3.561672133833392 + inference_time: 273579.0 + throughput: 3.655251316804287 estimated_peak_memory_range: - min: 7045120 - max: 344381152 + min: 7036928 + max: 344487936 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 32 - job_id: jep2ormrg + job_id: j0pxz1835 job_status: Passed torchscript_onnx_qnn: - inference_time: 315847.0 - throughput: 3.166089910621282 + inference_time: 283909.0 + throughput: 3.5222553705588764 estimated_peak_memory_range: - min: 3313664 - max: 83175488 + min: 2867200 + max: 79989328 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 52 - job_id: j7gj3r0vp + job_id: jw5601965 job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:39:17Z' + timestamp: '2024-08-26T23:56:42Z' - torchscript_onnx_tflite: - inference_time: 156719.0 - throughput: 6.380847248897709 + inference_time: 146996.0 + throughput: 6.802906201529293 estimated_peak_memory_range: - min: 6696960 - max: 463187240 + min: 6639616 + max: 462890008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 32 - job_id: jqpy8od8g + job_id: jo5mlz1dg job_status: Passed torchscript_onnx_qnn: - inference_time: 137761.0 - throughput: 7.258948468724821 + inference_time: 138201.0 + throughput: 7.235837656746333 estimated_peak_memory_range: - min: 10141696 - max: 11424960 + min: 10059776 + max: 11243560 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 52 - job_id: j1p3ozm3p + job_id: jogkkywwg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:39:13Z' + timestamp: '2024-08-26T23:56:39Z' - torchscript_onnx_tflite: - inference_time: 149646.0 - throughput: 6.682437218502332 + inference_time: 147262.0 + throughput: 6.790618082057829 estimated_peak_memory_range: - min: 6635520 - max: 547501624 + min: 6709248 + max: 9059976 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 32 - job_id: j2p0omr9p + job_id: jegnwedkg job_status: Passed torchscript_onnx_qnn: - inference_time: 138546.0 - throughput: 7.217819352417248 + inference_time: 138001.0 + throughput: 7.246324301997811 estimated_peak_memory_range: - min: 10100736 - max: 11296224 + min: 10121216 + max: 11704760 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 52 - job_id: jwgodlvq5 + job_id: jn5qd2xng job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:39:15Z' + timestamp: '2024-08-26T23:56:40Z' - torchscript_onnx_tflite: - inference_time: 153403.0 - throughput: 6.518777338122462 + inference_time: 150064.0 + throughput: 6.663823435334257 estimated_peak_memory_range: - min: 6672384 - max: 463215464 + min: 6696960 + max: 110701008 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 32 - job_id: j1p8je7k5 + job_id: jopr7ym0g job_status: Passed torchscript_onnx_qnn: - inference_time: 145904.0 - throughput: 6.853821690974888 + inference_time: 138239.0 + throughput: 7.233848624483684 estimated_peak_memory_range: - min: 10047488 - max: 11631984 + min: 10158080 + max: 11533600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 52 - job_id: j1pv2l1kg + job_id: j1glqk9jp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:39:16Z' + timestamp: '2024-08-26T23:56:41Z' - torchscript_onnx_qnn: - inference_time: 135742.0 - throughput: 7.366916650705014 + inference_time: 135550.0 + throughput: 7.377351530800443 estimated_peak_memory_range: min: 9850880 max: 9850880 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 52 - job_id: j1glwykjp + job_id: j2p0xr89p job_status: Passed torchscript_onnx: - inference_time: 147163.0 - throughput: 6.795186290032142 + inference_time: 147865.0 + throughput: 6.76292564163257 estimated_peak_memory_range: - min: 57704448 - max: 57704448 + min: 57827328 + max: 57827328 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 53 - job_id: jz5wyld3g + job_id: j1pvnw8kg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:39:20Z' + timestamp: '2024-08-26T23:56:44Z' diff --git a/qai_hub_models/models/vit/perf.yaml b/qai_hub_models/models/vit/perf.yaml index 8e93429a..c7556a90 100644 --- a/qai_hub_models/models/vit/perf.yaml +++ b/qai_hub_models/models/vit/perf.yaml @@ -45,26 +45,26 @@ models: - name: VIT performance_metrics: - torchscript_onnx_tflite: - inference_time: 139311.0 - throughput: 7.178184062995744 + inference_time: 66812.0 + throughput: 14.967371130934563 estimated_peak_memory_range: - min: 3760128 - max: 7442664 + min: 94208 + max: 3056040 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1567 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 1579 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1579 - job_id: jep2orlrg + job_id: jep2zmz4p job_status: Passed torchscript_onnx: - inference_time: 51185.0 - throughput: 19.536973722770345 + inference_time: 51087.0 + throughput: 19.574451425998788 estimated_peak_memory_range: - min: 110592 - max: 203453928 + min: 65536 + max: 202994464 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 976 - job_id: jmg9oz4wg + job_id: jmg9q47lp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,28 +81,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:38:18Z' + timestamp: '2024-08-26T23:55:45Z' - torchscript_onnx_tflite: - inference_time: 101467.0 - throughput: 9.855420974306918 + inference_time: 47669.0 + throughput: 20.97799408420567 estimated_peak_memory_range: - min: 3969024 - max: 387019344 + min: 65536 + max: 388677024 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1567 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 1579 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1579 - job_id: jqpy8o68g + job_id: jqpyydy7p job_status: Passed torchscript_onnx: - inference_time: 37669.0 - throughput: 26.54702806020866 + inference_time: 37633.0 + throughput: 26.57242313926607 estimated_peak_memory_range: min: 0 - max: 161074752 + max: 162040320 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 976 - job_id: jnp1on885 + job_id: jnp1m8k2p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,21 +119,21 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:38:19Z' + timestamp: '2024-08-26T23:55:46Z' - torchscript_onnx_tflite: - inference_time: 125738.0 - throughput: 7.953045221015127 + inference_time: 66903.0 + throughput: 14.94701283948403 estimated_peak_memory_range: - min: 3743744 - max: 7053968 + min: 90112 + max: 3413504 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1567 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 1579 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1579 - job_id: j2p0oml9p + job_id: j2p0xrx6p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,21 +142,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:38:03Z' + timestamp: '2024-08-26T23:55:33Z' - torchscript_onnx_tflite: - inference_time: 155225.0 - throughput: 6.442261233693026 + inference_time: 70073.0 + throughput: 14.270831846788349 estimated_peak_memory_range: - min: 3899392 - max: 373335472 + min: 331776 + max: 378557888 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1570 - layers_on_gpu: 1 - layers_on_cpu: 8 + layers_on_npu: 1579 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1579 - job_id: j1p8jezk5 + job_id: j1p8k7kxp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,21 +165,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:38:04Z' + timestamp: '2024-08-26T23:55:34Z' - torchscript_onnx_tflite: - inference_time: 132856.0 - throughput: 7.526946468356717 + inference_time: 66852.0 + throughput: 14.958415604619159 estimated_peak_memory_range: - min: 3657728 - max: 11792712 + min: 143360 + max: 2738192 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1567 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 1579 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1579 - job_id: jogk623w5 + job_id: jogkkyk2g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,21 +188,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:38:05Z' + timestamp: '2024-08-26T23:55:34Z' - torchscript_onnx_tflite: - inference_time: 122490.0 - throughput: 8.163931749530574 + inference_time: 66844.0 + throughput: 14.96020585243253 estimated_peak_memory_range: - min: 3670016 - max: 8784736 + min: 20480 + max: 2628472 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1567 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 1579 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1579 - job_id: jn5q4l3ng + job_id: jn5qd2d4g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,21 +211,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:38:06Z' + timestamp: '2024-08-26T23:55:35Z' - torchscript_onnx_tflite: - inference_time: 135214.0 - throughput: 7.39568387888828 + inference_time: 66903.0 + throughput: 14.94701283948403 estimated_peak_memory_range: - min: 3649536 - max: 12221328 + min: 188416 + max: 3048648 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 1567 - layers_on_gpu: 1 - layers_on_cpu: 11 + layers_on_npu: 1579 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 1579 - job_id: j1glwy3jp + job_id: j1glqkq8p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -234,13 +234,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:38:07Z' + timestamp: '2024-08-26T23:55:36Z' - torchscript_onnx: - inference_time: 52492.0 - throughput: 19.05052198430237 + inference_time: 52806.0 + throughput: 18.93724198007802 estimated_peak_memory_range: - min: 179519488 - max: 179519488 + min: 180502528 + max: 180502528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -248,7 +248,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 976 - job_id: jvgd6dvrp + job_id: jvgdmvyeg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -257,4 +257,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:38:20Z' + timestamp: '2024-08-26T23:55:47Z' diff --git a/qai_hub_models/models/whisper_base_en/export.py b/qai_hub_models/models/whisper_base_en/export.py index 742419c8..71e303a8 100644 --- a/qai_hub_models/models/whisper_base_en/export.py +++ b/qai_hub_models/models/whisper_base_en/export.py @@ -128,8 +128,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/whisper_base_en/perf.yaml b/qai_hub_models/models/whisper_base_en/perf.yaml index 3da18952..30a990eb 100644 --- a/qai_hub_models/models/whisper_base_en/perf.yaml +++ b/qai_hub_models/models/whisper_base_en/perf.yaml @@ -45,11 +45,11 @@ models: - name: WhisperEncoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 159738.0 - throughput: 6.2602511612765905 + inference_time: 159980.0 + throughput: 6.250781347668458 estimated_peak_memory_range: - min: 37089280 - max: 41581336 + min: 1208320 + max: 98257824 primary_compute_unit: GPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 419 layers_on_cpu: 0 total_layers: 419 - job_id: jogk627w5 + job_id: j1p8k7jxp job_status: Passed torchscript_onnx_qnn: - inference_time: 446209.0 - throughput: 2.2411022637373965 + inference_time: 419574.0 + throughput: 2.3833697988912563 estimated_peak_memory_range: - min: 499712 - max: 73329096 + min: 151552 + max: 92883984 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 531 - job_id: jz57oelvg + job_id: jz578dnlp job_status: Passed torchscript_onnx: - inference_time: 426191.0 - throughput: 2.3463658312822187 + inference_time: 433522.0 + throughput: 2.3066880112197303 estimated_peak_memory_range: - min: 1028096 - max: 136761368 + min: 12288 + max: 135253520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 380 - job_id: j1pv2lvkg + job_id: j1pvnwmjg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:37:18Z' + timestamp: '2024-08-26T23:54:50Z' - torchscript_onnx_tflite: - inference_time: 121877.0 - throughput: 8.204993559080057 + inference_time: 121356.0 + throughput: 8.240218860212927 estimated_peak_memory_range: - min: 35708928 - max: 78095072 + min: 37380096 + max: 79942112 primary_compute_unit: GPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 419 layers_on_cpu: 0 total_layers: 419 - job_id: j1glwy6jp + job_id: j1glqkw8p job_status: Passed torchscript_onnx_qnn: - inference_time: 314175.0 - throughput: 3.182939444577067 + inference_time: 317912.0 + throughput: 3.14552454767357 estimated_peak_memory_range: - min: 0 - max: 184935776 + min: 516096 + max: 190216928 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 531 - job_id: j0px0l63p + job_id: j0pxz1r15 job_status: Passed torchscript_onnx: - inference_time: 302435.0 - throughput: 3.306495610627077 + inference_time: 302223.0 + throughput: 3.308815014079008 estimated_peak_memory_range: - min: 0 - max: 847923472 + min: 98082816 + max: 947507600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 380 - job_id: jlpe67kog + job_id: jlpenvx1p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:37:20Z' + timestamp: '2024-08-26T23:54:51Z' - torchscript_onnx_tflite: - inference_time: 157407.0 - throughput: 6.352957619419721 + inference_time: 157859.0 + throughput: 6.3347671022874845 estimated_peak_memory_range: - min: 25391104 - max: 127140160 + min: 17833984 + max: 117603400 primary_compute_unit: GPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 419 layers_on_cpu: 0 total_layers: 419 - job_id: j1p3ozv3p + job_id: j1p3rmolp job_status: Passed torchscript_onnx_qnn: - inference_time: 454997.0 - throughput: 2.1978166889012454 + inference_time: 459220.0 + throughput: 2.1776055049867167 estimated_peak_memory_range: - min: 811008 - max: 2052832 + min: 307200 + max: 11400856 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 531 - job_id: jep2or9rg + job_id: jep2zmd4p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:37:07Z' + timestamp: '2024-08-26T23:54:41Z' - torchscript_onnx_tflite: - inference_time: 208402.0 - throughput: 4.798418441281753 + inference_time: 210075.0 + throughput: 4.760204688801618 estimated_peak_memory_range: - min: 37236736 - max: 85593424 + min: 37310464 + max: 86974992 primary_compute_unit: GPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 419 layers_on_cpu: 0 total_layers: 419 - job_id: j1pv2lrkg + job_id: j1pvnw2jg job_status: Passed torchscript_onnx_qnn: - inference_time: 461943.0 - throughput: 2.1647692464221775 + inference_time: 450251.0 + throughput: 2.220983407032966 estimated_peak_memory_range: min: 0 - max: 194536368 + max: 195291312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 531 - job_id: j1p3oz13p + job_id: j1p3rm7lp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:37:16Z' + timestamp: '2024-08-26T23:54:48Z' - torchscript_onnx_tflite: - inference_time: 159672.0 - throughput: 6.262838819580139 + inference_time: 160225.0 + throughput: 6.241223279762833 estimated_peak_memory_range: - min: 23068672 - max: 239580216 + min: 24608768 + max: 129714960 primary_compute_unit: GPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 419 layers_on_cpu: 0 total_layers: 419 - job_id: jlpe67wog + job_id: jlpenv61p job_status: Passed torchscript_onnx_qnn: - inference_time: 454778.0 - throughput: 2.198875055521595 + inference_time: 427238.0 + throughput: 2.3406157691965603 estimated_peak_memory_range: - min: 200704 - max: 21446416 + min: 176128 + max: 11278032 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 531 - job_id: j2p0om29p + job_id: j2p0xr96p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:37:09Z' + timestamp: '2024-08-26T23:54:43Z' - torchscript_onnx_tflite: - inference_time: 158027.0 - throughput: 6.328032551399445 + inference_time: 158892.0 + throughput: 6.293583062709262 estimated_peak_memory_range: - min: 12288 - max: 101438896 + min: 110592 + max: 4789768 primary_compute_unit: GPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 419 layers_on_cpu: 0 total_layers: 419 - job_id: jz5wyl33g + job_id: jz5wr9z6p job_status: Passed torchscript_onnx_qnn: - inference_time: 453327.0 - throughput: 2.2059131708457693 + inference_time: 429101.0 + throughput: 2.3304536694158253 estimated_peak_memory_range: - min: 819200 - max: 11857976 + min: 143360 + max: 11334768 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 531 - job_id: jogk62qw5 + job_id: jogkky02g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:37:11Z' + timestamp: '2024-08-26T23:54:45Z' - torchscript_onnx_tflite: - inference_time: 159199.0 - throughput: 6.281446491498063 + inference_time: 159232.0 + throughput: 6.280144694533762 estimated_peak_memory_range: min: 12288 - max: 97101352 + max: 100454760 primary_compute_unit: GPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 419 layers_on_cpu: 0 total_layers: 419 - job_id: jnp1onw85 + job_id: jnp1m812p job_status: Passed torchscript_onnx_qnn: - inference_time: 463764.0 - throughput: 2.15626913688859 + inference_time: 459137.0 + throughput: 2.1779991592923245 estimated_peak_memory_range: - min: 778240 - max: 2158368 + min: 737280 + max: 2338904 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 531 - job_id: j1glwy2jp + job_id: j1glqk88p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:37:14Z' + timestamp: '2024-08-26T23:54:46Z' - torchscript_onnx_qnn: - inference_time: 422662.0 - throughput: 2.3659567219196425 + inference_time: 423279.0 + throughput: 2.3625079439329615 estimated_peak_memory_range: min: 483328 max: 483328 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 531 - job_id: jegn1zmkp + job_id: jegnweqrg job_status: Passed torchscript_onnx: - inference_time: 385397.0 - throughput: 2.594726995799137 + inference_time: 388373.0 + throughput: 2.5748442862917864 estimated_peak_memory_range: - min: 139718656 - max: 139718656 + min: 139722752 + max: 139722752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 380 - job_id: jz5wylq3g + job_id: jz5wr9r6p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,30 +377,30 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:37:22Z' + timestamp: '2024-08-26T23:54:53Z' - name: WhisperDecoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 24476.0 - throughput: 40.85634907664651 + inference_time: 14149.0 + throughput: 70.67637288854336 estimated_peak_memory_range: - min: 43749376 - max: 46183416 + min: 4710400 + max: 7113280 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 976 - layers_on_gpu: 1 - layers_on_cpu: 6 + layers_on_npu: 983 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 983 - job_id: jn5q4leng + job_id: jn5qd244g job_status: Passed torchscript_onnx_qnn: - inference_time: 3995.0 - throughput: 250.31289111389236 + inference_time: 4042.0 + throughput: 247.40227610094013 estimated_peak_memory_range: - min: 19841024 - max: 30346192 + min: 21221376 + max: 31462744 primary_compute_unit: NPU precision: fp16 layer_info: @@ -408,14 +408,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 821 - job_id: jqp4eyd8g + job_id: jqp42w4vg job_status: Passed torchscript_onnx: - inference_time: 17252.0 - throughput: 57.964293994899144 + inference_time: 16845.0 + throughput: 59.364796675571384 estimated_peak_memory_range: - min: 42483712 - max: 45158304 + min: 61440 + max: 122238720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -423,7 +423,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 844 - job_id: j7gj3revp + job_id: j7gj8lyx5 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -432,28 +432,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:37:18Z' + timestamp: '2024-08-26T23:54:50Z' - torchscript_onnx_tflite: - inference_time: 20900.0 - throughput: 47.84688995215311 + inference_time: 11235.0 + throughput: 89.00756564307966 estimated_peak_memory_range: - min: 43696128 - max: 134642672 + min: 3743744 + max: 95044704 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 976 - layers_on_gpu: 1 - layers_on_cpu: 6 + layers_on_npu: 983 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 983 - job_id: jw56o8e65 + job_id: jw5601o05 job_status: Passed torchscript_onnx_qnn: - inference_time: 3072.0 - throughput: 325.5208333333333 + inference_time: 3038.0 + throughput: 329.1639236339697 estimated_peak_memory_range: - min: 85737472 - max: 122440448 + min: 19320832 + max: 59038768 primary_compute_unit: NPU precision: fp16 layer_info: @@ -461,14 +461,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 821 - job_id: jo5m906dg + job_id: jo5mlzkwg job_status: Passed torchscript_onnx: - inference_time: 13906.0 - throughput: 71.9114051488566 + inference_time: 13807.0 + throughput: 72.42702976750924 estimated_peak_memory_range: - min: 45133824 - max: 438459184 + min: 56381440 + max: 451842640 primary_compute_unit: NPU precision: fp16 layer_info: @@ -476,7 +476,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 844 - job_id: jygzzlrog + job_id: jygz07yk5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -485,28 +485,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:37:20Z' + timestamp: '2024-08-26T23:54:52Z' - torchscript_onnx_tflite: - inference_time: 24719.0 - throughput: 40.454710951090256 + inference_time: 13824.0 + throughput: 72.33796296296296 estimated_peak_memory_range: - min: 43753472 - max: 46318440 + min: 5754880 + max: 7623464 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 976 - layers_on_gpu: 1 - layers_on_cpu: 6 + layers_on_npu: 983 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 983 - job_id: jwgodlkq5 + job_id: jwgo9vdxg job_status: Passed torchscript_onnx_qnn: - inference_time: 4146.0 - throughput: 241.196333815726 + inference_time: 4055.0 + throughput: 246.6091245376079 estimated_peak_memory_range: - min: 21299200 - max: 22610696 + min: 19894272 + max: 22588912 primary_compute_unit: NPU precision: fp16 layer_info: @@ -514,7 +514,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 821 - job_id: jqpy8oj8g + job_id: jqpyyd27p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -523,28 +523,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:37:07Z' + timestamp: '2024-08-26T23:54:41Z' - torchscript_onnx_tflite: - inference_time: 32851.0 - throughput: 30.440473653770052 + inference_time: 26816.0 + throughput: 37.29116945107398 estimated_peak_memory_range: - min: 43831296 - max: 129252032 + min: 5783552 + max: 89221040 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 979 - layers_on_gpu: 1 - layers_on_cpu: 3 + layers_on_npu: 983 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 983 - job_id: j7gj3r2vp + job_id: j7gj8l3x5 job_status: Passed torchscript_onnx_qnn: - inference_time: 4969.0 - throughput: 201.2477359629704 + inference_time: 4834.0 + throughput: 206.8680182043856 estimated_peak_memory_range: min: 21213184 - max: 59577520 + max: 57762880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -552,7 +552,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 821 - job_id: jwgodlnq5 + job_id: jwgo9vwxg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -561,28 +561,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:37:16Z' + timestamp: '2024-08-26T23:54:48Z' - torchscript_onnx_tflite: - inference_time: 24673.0 - throughput: 40.530134154744054 + inference_time: 13959.0 + throughput: 71.63836951070994 estimated_peak_memory_range: - min: 35672064 - max: 44855440 + min: 5771264 + max: 8026216 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 976 - layers_on_gpu: 1 - layers_on_cpu: 6 + layers_on_npu: 983 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 983 - job_id: jygzzljog + job_id: jygz07zk5 job_status: Passed torchscript_onnx_qnn: - inference_time: 4061.0 - throughput: 246.2447672986949 + inference_time: 4048.0 + throughput: 247.03557312252966 estimated_peak_memory_range: - min: 21295104 - max: 22547888 + min: 19877888 + max: 21445288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -590,7 +590,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 821 - job_id: j1p8jemk5 + job_id: j1p8k7rxp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -599,28 +599,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:37:10Z' + timestamp: '2024-08-26T23:54:43Z' - torchscript_onnx_tflite: - inference_time: 25034.0 - throughput: 39.94567388351842 + inference_time: 14106.0 + throughput: 70.8918190840777 estimated_peak_memory_range: - min: 43692032 - max: 50050952 + min: 5750784 + max: 8511328 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 976 - layers_on_gpu: 1 - layers_on_cpu: 6 + layers_on_npu: 983 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 983 - job_id: jmg9ozywg + job_id: jmg9q42lp job_status: Passed torchscript_onnx_qnn: - inference_time: 4059.0 - throughput: 246.3661000246366 + inference_time: 4104.0 + throughput: 243.66471734892787 estimated_peak_memory_range: - min: 19873792 - max: 21282208 + min: 19857408 + max: 22601104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -628,7 +628,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 821 - job_id: jn5q4lrng + job_id: jn5qd214g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -637,28 +637,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:37:12Z' + timestamp: '2024-08-26T23:54:45Z' - torchscript_onnx_tflite: - inference_time: 24382.0 - throughput: 41.01386268558773 + inference_time: 14154.0 + throughput: 70.65140596297866 estimated_peak_memory_range: - min: 43843584 - max: 56551904 + min: 5758976 + max: 8166264 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 976 - layers_on_gpu: 1 - layers_on_cpu: 6 + layers_on_npu: 983 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 983 - job_id: jvgd6dqrp + job_id: jvgdmv4eg job_status: Passed torchscript_onnx_qnn: - inference_time: 4016.0 - throughput: 249.003984063745 + inference_time: 4031.0 + throughput: 248.07740014884644 estimated_peak_memory_range: - min: 18489344 - max: 19849440 + min: 19869696 + max: 25403608 primary_compute_unit: NPU precision: fp16 layer_info: @@ -666,7 +666,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 821 - job_id: jw56o8z65 + job_id: jw5601m05 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -675,10 +675,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:37:14Z' + timestamp: '2024-08-26T23:54:47Z' - torchscript_onnx_qnn: - inference_time: 3625.0 - throughput: 275.86206896551727 + inference_time: 4047.0 + throughput: 247.09661477637755 estimated_peak_memory_range: min: 21229568 max: 21229568 @@ -689,14 +689,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 821 - job_id: joprxl20p + job_id: jopr7yd9g job_status: Passed torchscript_onnx: - inference_time: 14348.0 - throughput: 69.69612489545581 + inference_time: 14352.0 + throughput: 69.67670011148272 estimated_peak_memory_range: - min: 113352704 - max: 113352704 + min: 113385472 + max: 113385472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -704,7 +704,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 844 - job_id: jmg9ozwwg + job_id: jmg9q4qlp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -713,4 +713,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:37:22Z' + timestamp: '2024-08-26T23:54:53Z' diff --git a/qai_hub_models/models/whisper_small_en/export.py b/qai_hub_models/models/whisper_small_en/export.py index 35de9458..b626a255 100644 --- a/qai_hub_models/models/whisper_small_en/export.py +++ b/qai_hub_models/models/whisper_small_en/export.py @@ -128,8 +128,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/whisper_small_en/perf.yaml b/qai_hub_models/models/whisper_small_en/perf.yaml index 0099d35e..d0304b2f 100644 --- a/qai_hub_models/models/whisper_small_en/perf.yaml +++ b/qai_hub_models/models/whisper_small_en/perf.yaml @@ -45,11 +45,11 @@ models: - name: WhisperEncoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 623350.0 - throughput: 1.6042351808775166 + inference_time: 620742.0 + throughput: 1.6109752521981757 estimated_peak_memory_range: - min: 102223872 - max: 556720864 + min: 20480 + max: 435146928 primary_compute_unit: GPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 911 layers_on_cpu: 0 total_layers: 911 - job_id: jlpe67d1g + job_id: j1p3rm9zp job_status: Passed torchscript_onnx_qnn: - inference_time: 1890126.0 - throughput: 0.5290652580833235 + inference_time: 1884836.0 + throughput: 0.530550138049146 estimated_peak_memory_range: - min: 86016 - max: 254815320 + min: 73728 + max: 261705704 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1329 - job_id: j2p0om46p + job_id: jz578delp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,13 +81,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:35:22Z' + timestamp: '2024-08-26T23:53:02Z' - torchscript_onnx_tflite: - inference_time: 462336.0 - throughput: 2.1629291251384273 + inference_time: 481840.0 + throughput: 2.0753777187448117 estimated_peak_memory_range: - min: 111874048 - max: 204887840 + min: 111771648 + max: 205456544 primary_compute_unit: GPU precision: fp16 layer_info: @@ -95,14 +95,14 @@ models: layers_on_gpu: 911 layers_on_cpu: 0 total_layers: 911 - job_id: jz5wylw6g + job_id: j1pvnwdmg job_status: Passed torchscript_onnx_qnn: - inference_time: 1404578.0 - throughput: 0.7119576128915589 + inference_time: 1412053.0 + throughput: 0.7081887152960973 estimated_peak_memory_range: - min: 659456 - max: 542879200 + min: 495616 + max: 546518912 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1329 - job_id: jogk62v25 + job_id: j0pxz1l15 job_status: Passed torchscript_onnx: - inference_time: 1183349.0 - throughput: 0.8450592344270371 + inference_time: 1187219.0 + throughput: 0.8423045790203829 estimated_peak_memory_range: - min: 55115776 - max: 3115065952 + min: 79192064 + max: 3140023472 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,7 +125,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 884 - job_id: jnp1ony85 + job_id: jlpenv71p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -134,13 +134,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:35:42Z' + timestamp: '2024-08-26T23:53:19Z' - torchscript_onnx_tflite: - inference_time: 610414.0 - throughput: 1.638232412755933 + inference_time: 612400.0 + throughput: 1.6329196603527107 estimated_peak_memory_range: - min: 0 - max: 452809480 + min: 114688 + max: 406610272 primary_compute_unit: GPU precision: fp16 layer_info: @@ -148,14 +148,14 @@ models: layers_on_gpu: 911 layers_on_cpu: 0 total_layers: 911 - job_id: jnp1on225 + job_id: jlpenvz0p job_status: Passed torchscript_onnx_qnn: - inference_time: 1279192.0 - throughput: 0.7817434755689529 + inference_time: 1229024.0 + throughput: 0.813653761033145 estimated_peak_memory_range: - min: 974848 - max: 2210560 + min: 978944 + max: 2286296 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,7 +163,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1329 - job_id: j1p3oznlp + job_id: jep2zmr4p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -172,13 +172,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:35:29Z' + timestamp: '2024-08-26T23:53:08Z' - torchscript_onnx_tflite: - inference_time: 829351.0 - throughput: 1.2057620959039055 + inference_time: 839639.0 + throughput: 1.1909880317612689 estimated_peak_memory_range: - min: 110411776 - max: 210805296 + min: 110407680 + max: 210071696 primary_compute_unit: GPU precision: fp16 layer_info: @@ -186,7 +186,7 @@ models: layers_on_gpu: 911 layers_on_cpu: 0 total_layers: 911 - job_id: jz57oe2lg + job_id: jz5wr97jp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -195,13 +195,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:35:13Z' + timestamp: '2024-08-26T23:52:54Z' - torchscript_onnx_tflite: - inference_time: 613595.0 - throughput: 1.62973948614314 + inference_time: 612849.0 + throughput: 1.6317233119414407 estimated_peak_memory_range: - min: 2187264 - max: 322938640 + min: 0 + max: 439066016 primary_compute_unit: GPU precision: fp16 layer_info: @@ -209,14 +209,14 @@ models: layers_on_gpu: 911 layers_on_cpu: 0 total_layers: 911 - job_id: j0px0l91p + job_id: jnp1m8jlp job_status: Passed torchscript_onnx_qnn: - inference_time: 1269373.0 - throughput: 0.7877905075970577 + inference_time: 1253639.0 + throughput: 0.7976778003875119 estimated_peak_memory_range: - min: 749568 - max: 1998944 + min: 774144 + max: 2033976 primary_compute_unit: NPU precision: fp16 layer_info: @@ -224,7 +224,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1329 - job_id: j1pv2lqjg + job_id: j2p0xrm6p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -233,13 +233,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:35:31Z' + timestamp: '2024-08-26T23:53:10Z' - torchscript_onnx_tflite: - inference_time: 610375.0 - throughput: 1.6383370878558263 + inference_time: 612352.0 + throughput: 1.6330476588628762 estimated_peak_memory_range: - min: 16384 - max: 444125352 + min: 57819136 + max: 502907864 primary_compute_unit: GPU precision: fp16 layer_info: @@ -247,14 +247,14 @@ models: layers_on_gpu: 911 layers_on_cpu: 0 total_layers: 911 - job_id: jegn1z0rp + job_id: jz5wr9l6p job_status: Passed torchscript_onnx_qnn: - inference_time: 1245092.0 - throughput: 0.803153501909899 + inference_time: 1252449.0 + throughput: 0.79843570476722 estimated_peak_memory_range: - min: 802816 - max: 1929752 + min: 937984 + max: 2360504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -262,7 +262,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1329 - job_id: jlpe67o1g + job_id: jogkky22g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -271,13 +271,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:35:34Z' + timestamp: '2024-08-26T23:53:12Z' - torchscript_onnx_tflite: - inference_time: 610977.0 - throughput: 1.636722822626711 + inference_time: 614352.0 + throughput: 1.627731333177071 estimated_peak_memory_range: - min: 49229824 - max: 495393512 + min: 111796224 + max: 119975208 primary_compute_unit: GPU precision: fp16 layer_info: @@ -285,14 +285,14 @@ models: layers_on_gpu: 911 layers_on_cpu: 0 total_layers: 911 - job_id: jep2orx4g + job_id: jnp1m8n2p job_status: Passed torchscript_onnx_qnn: - inference_time: 1294694.0 - throughput: 0.7723832813004463 + inference_time: 1238815.0 + throughput: 0.8072230316875401 estimated_peak_memory_range: - min: 958464 - max: 2351384 + min: 659456 + max: 2080520 primary_compute_unit: NPU precision: fp16 layer_info: @@ -300,7 +300,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1329 - job_id: jz5wyl26g + job_id: j1glqky8p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -309,10 +309,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:35:36Z' + timestamp: '2024-08-26T23:53:14Z' - torchscript_onnx_qnn: - inference_time: 1087181.0 - throughput: 0.9198100408303677 + inference_time: 1116556.0 + throughput: 0.8956111471345817 estimated_peak_memory_range: min: 483328 max: 483328 @@ -323,14 +323,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 1329 - job_id: j1glwy48p + job_id: jegnwezrg job_status: Passed torchscript_onnx: - inference_time: 1506275.0 - throughput: 0.663889396026622 + inference_time: 1537750.0 + throughput: 0.6503007641033979 estimated_peak_memory_range: - min: 454066176 - max: 454066176 + min: 455184384 + max: 455184384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -338,7 +338,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 884 - job_id: jz57oe0vg + job_id: jz5wr9y6p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -347,30 +347,30 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:35:45Z' + timestamp: '2024-08-26T23:53:21Z' - name: WhisperDecoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 54275.0 - throughput: 18.424689083371717 + inference_time: 25837.0 + throughput: 38.704183922282 estimated_peak_memory_range: - min: 128020480 - max: 131109392 + min: 16752640 + max: 20071120 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 2560 - layers_on_gpu: 1 - layers_on_cpu: 12 + layers_on_npu: 2573 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 2573 - job_id: jygzzl3kg + job_id: jwgo9vrdg job_status: Passed torchscript_onnx_qnn: - inference_time: 11949.0 - throughput: 83.68901163277262 + inference_time: 11849.0 + throughput: 84.39530762089628 estimated_peak_memory_range: - min: 63619072 - max: 133879384 + min: 64827392 + max: 134580120 primary_compute_unit: NPU precision: fp16 layer_info: @@ -378,14 +378,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2255 - job_id: j1p8je2x5 + job_id: jqp42wyvg job_status: Passed torchscript_onnx: - inference_time: 61557.0 - throughput: 16.245106161768767 + inference_time: 62734.0 + throughput: 15.940319444001657 estimated_peak_memory_range: - min: 163840 - max: 271544440 + min: 77824 + max: 271541632 primary_compute_unit: NPU precision: fp16 layer_info: @@ -393,7 +393,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2302 - job_id: jmg9ozjwg + job_id: j7gj8lrx5 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -402,28 +402,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:35:41Z' + timestamp: '2024-08-26T23:53:18Z' - torchscript_onnx_tflite: - inference_time: 48761.0 - throughput: 20.508193023112735 + inference_time: 19207.0 + throughput: 52.06435153850159 estimated_peak_memory_range: - min: 128675840 - max: 759930960 + min: 15691776 + max: 1155777664 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 2560 - layers_on_gpu: 1 - layers_on_cpu: 12 + layers_on_npu: 2573 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 2573 - job_id: jmg9oz0lg + job_id: j7gj8l785 job_status: Passed torchscript_onnx_qnn: - inference_time: 9326.0 - throughput: 107.2271070126528 + inference_time: 9587.0 + throughput: 104.30791697089809 estimated_peak_memory_range: - min: 55336960 - max: 147292400 + min: 52359168 + max: 144777952 primary_compute_unit: NPU precision: fp16 layer_info: @@ -431,14 +431,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2255 - job_id: jn5q4l04g + job_id: jo5mlz0wg job_status: Passed torchscript_onnx: - inference_time: 48370.0 - throughput: 20.673971469919373 + inference_time: 48157.0 + throughput: 20.76541312789418 estimated_peak_memory_range: - min: 96161792 - max: 1506791040 + min: 64299008 + max: 1478360880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -446,7 +446,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2302 - job_id: jvgd6derp + job_id: jygz07lk5 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -455,28 +455,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:35:43Z' + timestamp: '2024-08-26T23:53:20Z' - torchscript_onnx_tflite: - inference_time: 57007.0 - throughput: 17.541705404599433 + inference_time: 25348.0 + throughput: 39.45084424806691 estimated_peak_memory_range: - min: 125976576 - max: 132614968 + min: 15568896 + max: 18687880 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 2560 - layers_on_gpu: 1 - layers_on_cpu: 12 + layers_on_npu: 2573 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 2573 - job_id: jvgd6dnep + job_id: jygz07m65 job_status: Passed torchscript_onnx_qnn: - inference_time: 12157.0 - throughput: 82.25713580653121 + inference_time: 12002.0 + throughput: 83.31944675887353 estimated_peak_memory_range: - min: 63676416 - max: 64868232 + min: 63713280 + max: 64975920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -484,7 +484,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2255 - job_id: jwgodlzx5 + job_id: jqpyydo7p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -493,28 +493,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:35:29Z' + timestamp: '2024-08-26T23:53:08Z' - torchscript_onnx_tflite: - inference_time: 64921.0 - throughput: 15.403336362656152 + inference_time: 27400.0 + throughput: 36.496350364963504 estimated_peak_memory_range: - min: 128880640 - max: 742622848 + min: 16547840 + max: 1139614976 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 2563 - layers_on_gpu: 1 - layers_on_cpu: 9 + layers_on_npu: 2573 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 2573 - job_id: jqp4eynvg + job_id: jmg9q4mvp job_status: Passed torchscript_onnx_qnn: - inference_time: 15534.0 - throughput: 64.37491953135059 + inference_time: 15752.0 + throughput: 63.48400203148807 estimated_peak_memory_range: - min: 58114048 - max: 156905632 + min: 55336960 + max: 152070064 primary_compute_unit: NPU precision: fp16 layer_info: @@ -522,7 +522,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2255 - job_id: jvgd6deep + job_id: jwgo9vlxg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -531,28 +531,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:35:39Z' + timestamp: '2024-08-26T23:53:16Z' - torchscript_onnx_tflite: - inference_time: 56992.0 - throughput: 17.54632229084784 + inference_time: 25599.0 + throughput: 39.064025938513225 estimated_peak_memory_range: - min: 125800448 - max: 133750672 + min: 15675392 + max: 18542328 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 2560 - layers_on_gpu: 1 - layers_on_cpu: 12 + layers_on_npu: 2573 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 2573 - job_id: jo5m90ewg + job_id: jvgdmv3lg job_status: Passed torchscript_onnx_qnn: inference_time: 12230.0 throughput: 81.76614881439085 estimated_peak_memory_range: - min: 63721472 - max: 65090320 + min: 63705088 + max: 65004736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -560,7 +560,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2255 - job_id: j7gj3rdxp + job_id: j1p8k7exp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -569,28 +569,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:35:32Z' + timestamp: '2024-08-26T23:53:10Z' - torchscript_onnx_tflite: - inference_time: 56508.0 - throughput: 17.69660932965244 + inference_time: 25403.0 + throughput: 39.365429280006296 estimated_peak_memory_range: - min: 129130496 - max: 138368720 + min: 16805888 + max: 19708104 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 2560 - layers_on_gpu: 1 - layers_on_cpu: 12 + layers_on_npu: 2573 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 2573 - job_id: joprxl69p + job_id: jmg9q4zlp job_status: Passed torchscript_onnx_qnn: - inference_time: 11899.0 - throughput: 84.04067568703252 + inference_time: 12094.0 + throughput: 82.6856292376385 estimated_peak_memory_range: - min: 67727360 - max: 68977336 + min: 63655936 + max: 73336792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -598,7 +598,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2255 - job_id: jygzzl2kg + job_id: jn5qd2l4g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -607,28 +607,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:35:34Z' + timestamp: '2024-08-26T23:53:12Z' - torchscript_onnx_tflite: - inference_time: 57362.0 - throughput: 17.433143893169696 + inference_time: 25545.0 + throughput: 39.146604032100214 estimated_peak_memory_range: - min: 129142784 - max: 132570824 + min: 16809984 + max: 19577200 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 2560 - layers_on_gpu: 1 - layers_on_cpu: 12 + layers_on_npu: 2573 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 2573 - job_id: jqpy8oz7g + job_id: jvgdmvdeg job_status: Passed torchscript_onnx_qnn: - inference_time: 12427.0 - throughput: 80.46994447573832 + inference_time: 12719.0 + throughput: 78.62253321802028 estimated_peak_memory_range: - min: 63705088 - max: 73929240 + min: 63729664 + max: 65150992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -636,7 +636,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2255 - job_id: jmg9ozjlg + job_id: jw5601805 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -645,13 +645,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:35:36Z' + timestamp: '2024-08-26T23:53:14Z' - torchscript_onnx_qnn: - inference_time: 10684.0 - throughput: 93.59790340696368 + inference_time: 10325.0 + throughput: 96.85230024213075 estimated_peak_memory_range: - min: 63696896 - max: 63696896 + min: 63700992 + max: 63700992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -659,14 +659,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2255 - job_id: jw56o8205 + job_id: jopr7yl9g job_status: Passed torchscript_onnx: - inference_time: 53768.0 - throughput: 18.598422853742004 + inference_time: 53081.0 + throughput: 18.839132646332963 estimated_peak_memory_range: - min: 243802112 - max: 243802112 + min: 243683328 + max: 243683328 primary_compute_unit: NPU precision: fp16 layer_info: @@ -674,7 +674,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 2302 - job_id: jqp4eyk8g + job_id: jmg9q4olp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -683,4 +683,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:35:45Z' + timestamp: '2024-08-26T23:53:21Z' diff --git a/qai_hub_models/models/whisper_tiny_en/export.py b/qai_hub_models/models/whisper_tiny_en/export.py index 58d19435..c64fd947 100644 --- a/qai_hub_models/models/whisper_tiny_en/export.py +++ b/qai_hub_models/models/whisper_tiny_en/export.py @@ -128,8 +128,9 @@ def export_model( compile_jobs: Dict[str, hub.client.CompileJob] = {} for component_name, component in components_dict.items(): - # Trace the model input_spec = component.get_input_spec() + + # Trace the model source_model = torch.jit.trace( component.to("cpu"), make_torch_inputs(input_spec) ) diff --git a/qai_hub_models/models/whisper_tiny_en/perf.yaml b/qai_hub_models/models/whisper_tiny_en/perf.yaml index 14e30f47..30d41c23 100644 --- a/qai_hub_models/models/whisper_tiny_en/perf.yaml +++ b/qai_hub_models/models/whisper_tiny_en/perf.yaml @@ -45,11 +45,11 @@ models: - name: WhisperEncoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 70794.0 - throughput: 14.125490860807414 + inference_time: 70172.0 + throughput: 14.250698284215927 estimated_peak_memory_range: - min: 12288 - max: 129479944 + min: 1101824 + max: 141729832 primary_compute_unit: GPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 271 layers_on_cpu: 0 total_layers: 271 - job_id: jnp1onz25 + job_id: jlpenvl0p job_status: Passed torchscript_onnx_qnn: - inference_time: 277614.0 - throughput: 3.602123812199673 + inference_time: 277593.0 + throughput: 3.6023963140280917 estimated_peak_memory_range: - min: 503808 - max: 45083008 + min: 143360 + max: 53775656 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +72,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 313 - job_id: j1glwym8p + job_id: j2p0xr6ep job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,13 +81,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:33:32Z' + timestamp: '2024-08-26T23:51:20Z' - torchscript_onnx_tflite: - inference_time: 53893.0 - throughput: 18.555285473067002 + inference_time: 53819.0 + throughput: 18.580798602723945 estimated_peak_memory_range: - min: 18890752 - max: 49918304 + min: 18829312 + max: 48561056 primary_compute_unit: GPU precision: fp16 layer_info: @@ -95,14 +95,14 @@ models: layers_on_gpu: 271 layers_on_cpu: 0 total_layers: 271 - job_id: jz57oerlg + job_id: jz5wr91jp job_status: Passed torchscript_onnx_qnn: - inference_time: 208826.0 - throughput: 4.788675739610968 + inference_time: 207316.0 + throughput: 4.823554380752088 estimated_peak_memory_range: - min: 28672 - max: 128889984 + min: 32768 + max: 131420096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +110,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 313 - job_id: j1p3oz0lp + job_id: jogkky8og job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,13 +119,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:33:34Z' + timestamp: '2024-08-26T23:51:21Z' - torchscript_onnx_tflite: - inference_time: 68484.0 - throughput: 14.601950820629636 + inference_time: 68332.0 + throughput: 14.634431891353978 estimated_peak_memory_range: - min: 8667136 - max: 140253200 + min: 6893568 + max: 132899192 primary_compute_unit: GPU precision: fp16 layer_info: @@ -133,14 +133,14 @@ models: layers_on_gpu: 271 layers_on_cpu: 0 total_layers: 271 - job_id: j0px0lo1p + job_id: jnp1m8vlp job_status: Passed torchscript_onnx_qnn: - inference_time: 245424.0 - throughput: 4.074581133059522 + inference_time: 246069.0 + throughput: 4.0639007757986585 estimated_peak_memory_range: - min: 147456 - max: 10827024 + min: 675840 + max: 2015088 primary_compute_unit: NPU precision: fp16 layer_info: @@ -148,7 +148,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 313 - job_id: jlpe67m1g + job_id: j1p3rm6zp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -157,13 +157,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:33:39Z' + timestamp: '2024-08-26T23:51:25Z' - torchscript_onnx_tflite: - inference_time: 90199.0 - throughput: 11.086597412388164 + inference_time: 91965.0 + throughput: 10.873701951829501 estimated_peak_memory_range: - min: 18464768 - max: 53249440 + min: 18673664 + max: 54974384 primary_compute_unit: GPU precision: fp16 layer_info: @@ -171,14 +171,14 @@ models: layers_on_gpu: 271 layers_on_cpu: 0 total_layers: 271 - job_id: jegn1zvrp + job_id: jz578d9rp job_status: Passed torchscript_onnx_qnn: - inference_time: 296424.0 - throughput: 3.373546001673279 + inference_time: 302129.0 + throughput: 3.3098444704083354 estimated_peak_memory_range: min: 0 - max: 133059056 + max: 132336000 primary_compute_unit: NPU precision: fp16 layer_info: @@ -186,7 +186,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 313 - job_id: j0px0lk1p + job_id: jnp1m86lp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -195,13 +195,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:33:48Z' + timestamp: '2024-08-26T23:51:32Z' - torchscript_onnx_tflite: - inference_time: 70185.0 - throughput: 14.248058702001853 + inference_time: 69535.0 + throughput: 14.381246854102251 estimated_peak_memory_range: - min: 15478784 - max: 98540592 + min: 3887104 + max: 164282792 primary_compute_unit: GPU precision: fp16 layer_info: @@ -209,14 +209,14 @@ models: layers_on_gpu: 271 layers_on_cpu: 0 total_layers: 271 - job_id: jep2ory4g + job_id: j0pxz1x95 job_status: Passed torchscript_onnx_qnn: - inference_time: 251808.0 - throughput: 3.971279705172195 + inference_time: 251118.0 + throughput: 3.9821916389904346 estimated_peak_memory_range: - min: 786432 - max: 2092976 + min: 675840 + max: 2198504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -224,7 +224,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 313 - job_id: jz5wyle6g + job_id: j1pvnw7mg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -233,13 +233,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:33:41Z' + timestamp: '2024-08-26T23:51:27Z' - torchscript_onnx_tflite: - inference_time: 68687.0 - throughput: 14.558795696419992 + inference_time: 69120.0 + throughput: 14.467592592592593 estimated_peak_memory_range: - min: 6557696 - max: 53684224 + min: 11776000 + max: 58884712 primary_compute_unit: GPU precision: fp16 layer_info: @@ -247,14 +247,14 @@ models: layers_on_gpu: 271 layers_on_cpu: 0 total_layers: 271 - job_id: j2p0om06p + job_id: jegnwekmg job_status: Passed torchscript_onnx_qnn: - inference_time: 247610.0 - throughput: 4.0386091030249185 + inference_time: 253876.0 + throughput: 3.938930816619137 estimated_peak_memory_range: - min: 749568 - max: 2033168 + min: 241664 + max: 6004800 primary_compute_unit: NPU precision: fp16 layer_info: @@ -262,7 +262,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 313 - job_id: jnp1on425 + job_id: jlpenvy0p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -271,13 +271,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:33:43Z' + timestamp: '2024-08-26T23:51:29Z' - torchscript_onnx_tflite: - inference_time: 69921.0 - throughput: 14.30185495058709 + inference_time: 68749.0 + throughput: 14.54566611877991 estimated_peak_memory_range: - min: 11456512 - max: 116395544 + min: 16384 + max: 148749024 primary_compute_unit: GPU precision: fp16 layer_info: @@ -285,14 +285,14 @@ models: layers_on_gpu: 271 layers_on_cpu: 0 total_layers: 271 - job_id: jogk62x25 + job_id: jep2zmemp job_status: Passed torchscript_onnx_qnn: - inference_time: 245364.0 - throughput: 4.075577509333073 + inference_time: 250173.0 + throughput: 3.9972339141314213 estimated_peak_memory_range: - min: 184320 - max: 1407520 + min: 90112 + max: 16173392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -300,7 +300,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 313 - job_id: jz57oeylg + job_id: jz5wr94jp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -309,13 +309,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:33:45Z' + timestamp: '2024-08-26T23:51:30Z' - torchscript_onnx_qnn: - inference_time: 239738.0 - throughput: 4.171220248771576 + inference_time: 236089.0 + throughput: 4.23569077763047 estimated_peak_memory_range: - min: 491520 - max: 491520 + min: 495616 + max: 495616 primary_compute_unit: NPU precision: fp16 layer_info: @@ -323,7 +323,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 313 - job_id: j1pv2lkjg + job_id: j1glqkllp job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -332,30 +332,30 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:33:37Z' + timestamp: '2024-08-26T23:51:23Z' - name: WhisperDecoder performance_metrics: - torchscript_onnx_tflite: - inference_time: 10440.0 - throughput: 95.78544061302682 + inference_time: 3703.0 + throughput: 270.0513097488523 estimated_peak_memory_range: - min: 21962752 - max: 29655192 + min: 2981888 + max: 5076456 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 552 - layers_on_gpu: 1 - layers_on_cpu: 4 + layers_on_npu: 557 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 557 - job_id: jvgd6d1ep + job_id: jygz07465 job_status: Passed torchscript_onnx_qnn: - inference_time: 2237.0 - throughput: 447.02726866338844 + inference_time: 2242.0 + throughput: 446.03033006244425 estimated_peak_memory_range: - min: 12288 - max: 139694088 + min: 1409024 + max: 161268480 primary_compute_unit: NPU precision: fp16 layer_info: @@ -363,14 +363,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 447 - job_id: jw56o8405 + job_id: j1p8k718p job_status: Passed torchscript_onnx: - inference_time: 5353.0 - throughput: 186.81113394358303 + inference_time: 5310.0 + throughput: 188.32391713747646 estimated_peak_memory_range: - min: 24576 - max: 78988448 + min: 16384 + max: 79332144 primary_compute_unit: NPU precision: fp16 layer_info: @@ -378,7 +378,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 462 - job_id: joprxlv9p + job_id: jqp42w1lg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -387,28 +387,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:33:50Z' + timestamp: '2024-08-26T23:51:34Z' - torchscript_onnx_tflite: - inference_time: 8883.0 - throughput: 112.57458065968704 + inference_time: 2777.0 + throughput: 360.1008282319049 estimated_peak_memory_range: - min: 20967424 - max: 97370000 + min: 933888 + max: 226456928 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 552 - layers_on_gpu: 1 - layers_on_cpu: 4 + layers_on_npu: 557 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 557 - job_id: jqp4eyrvg + job_id: jmg9q4xvp job_status: Passed torchscript_onnx_qnn: - inference_time: 1840.0 - throughput: 543.4782608695652 + inference_time: 1600.0 + throughput: 625.0 estimated_peak_memory_range: - min: 0 - max: 21943360 + min: 4624384 + max: 27615920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -416,14 +416,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 447 - job_id: jwgodl6x5 + job_id: jn5qd2vmg job_status: Passed torchscript_onnx: - inference_time: 4181.0 - throughput: 239.17723032767282 + inference_time: 4060.0 + throughput: 246.30541871921181 estimated_peak_memory_range: - min: 0 - max: 392119520 + min: 28516352 + max: 423256560 primary_compute_unit: NPU precision: fp16 layer_info: @@ -431,7 +431,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 462 - job_id: jqpy8o17g + job_id: jo5mlzmqg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -440,28 +440,28 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:33:52Z' + timestamp: '2024-08-26T23:51:36Z' - torchscript_onnx_tflite: - inference_time: 10309.0 - throughput: 97.00261907071491 + inference_time: 3676.0 + throughput: 272.0348204570185 estimated_peak_memory_range: - min: 21958656 - max: 29014624 + min: 3125248 + max: 5167296 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 552 - layers_on_gpu: 1 - layers_on_cpu: 4 + layers_on_npu: 557 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 557 - job_id: jo5m90xwg + job_id: jvgdmvzlg job_status: Passed torchscript_onnx_qnn: - inference_time: 2250.0 - throughput: 444.44444444444446 + inference_time: 2187.0 + throughput: 457.2473708276177 estimated_peak_memory_range: - min: 10665984 - max: 12390216 + min: 10690560 + max: 12037704 primary_compute_unit: NPU precision: fp16 layer_info: @@ -469,7 +469,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 447 - job_id: jygzzldkg + job_id: jwgo9v8dg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -478,28 +478,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:33:39Z' + timestamp: '2024-08-26T23:51:25Z' - torchscript_onnx_tflite: - inference_time: 16337.0 - throughput: 61.21074860745547 + inference_time: 4311.0 + throughput: 231.9647413593134 estimated_peak_memory_range: - min: 17195008 - max: 89354064 + min: 2977792 + max: 224873136 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 555 - layers_on_gpu: 1 - layers_on_cpu: 1 + layers_on_npu: 557 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 557 - job_id: joprxl39p + job_id: jqp42w3lg job_status: Passed torchscript_onnx_qnn: - inference_time: 2631.0 - throughput: 380.08361839604714 + inference_time: 2645.0 + throughput: 378.0718336483932 estimated_peak_memory_range: - min: 4624384 - max: 29664832 + min: 10620928 + max: 35347568 primary_compute_unit: NPU precision: fp16 layer_info: @@ -507,7 +507,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 447 - job_id: jo5m90nwg + job_id: jvgdmv2lg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -516,28 +516,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:33:49Z' + timestamp: '2024-08-26T23:51:33Z' - torchscript_onnx_tflite: - inference_time: 10503.0 - throughput: 95.21089212605922 + inference_time: 3691.0 + throughput: 270.92928745597396 estimated_peak_memory_range: - min: 21970944 - max: 28886856 + min: 2977792 + max: 5224808 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 552 - layers_on_gpu: 1 - layers_on_cpu: 4 + layers_on_npu: 557 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 557 - job_id: jqpy8o37g + job_id: jo5mlz8qg job_status: Passed torchscript_onnx_qnn: - inference_time: 2219.0 - throughput: 450.6534474988734 + inference_time: 2249.0 + throughput: 444.642063139173 estimated_peak_memory_range: min: 10674176 - max: 11872192 + max: 13100128 primary_compute_unit: NPU precision: fp16 layer_info: @@ -545,7 +545,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 447 - job_id: jmg9ozllg + job_id: j7gj8lq85 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -554,28 +554,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:33:41Z' + timestamp: '2024-08-26T23:51:27Z' - torchscript_onnx_tflite: - inference_time: 10218.0 - throughput: 97.86651008025053 + inference_time: 3736.0 + throughput: 267.6659528907923 estimated_peak_memory_range: - min: 21958656 - max: 26807496 + min: 2973696 + max: 4914256 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 552 - layers_on_gpu: 1 - layers_on_cpu: 4 + layers_on_npu: 557 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 557 - job_id: j1p8jeyx5 + job_id: jopr7yweg job_status: Passed torchscript_onnx_qnn: - inference_time: 2208.0 - throughput: 452.8985507246377 + inference_time: 2140.0 + throughput: 467.2897196261682 estimated_peak_memory_range: - min: 10670080 - max: 11839224 + min: 4657152 + max: 6261392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -583,7 +583,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 447 - job_id: jvgd6dxep + job_id: jygz07n65 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -592,28 +592,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:33:44Z' + timestamp: '2024-08-26T23:51:29Z' - torchscript_onnx_tflite: - inference_time: 10286.0 - throughput: 97.21952167995333 + inference_time: 3758.0 + throughput: 266.0989888238425 estimated_peak_memory_range: - min: 22003712 - max: 24753056 + min: 2985984 + max: 4999728 primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 552 - layers_on_gpu: 1 - layers_on_cpu: 4 + layers_on_npu: 557 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 557 - job_id: jn5q4lq4g + job_id: jqpyydm4p job_status: Passed torchscript_onnx_qnn: - inference_time: 2143.0 - throughput: 466.63555762949136 + inference_time: 2175.0 + throughput: 459.7701149425287 estimated_peak_memory_range: - min: 5685248 - max: 6933488 + min: 10670080 + max: 11955584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -621,7 +621,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 447 - job_id: jqp4eylvg + job_id: jmg9q4dvp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -630,10 +630,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:33:46Z' + timestamp: '2024-08-26T23:51:31Z' - torchscript_onnx_qnn: - inference_time: 2056.0 - throughput: 486.38132295719845 + inference_time: 2108.0 + throughput: 474.3833017077799 estimated_peak_memory_range: min: 10629120 max: 10629120 @@ -644,14 +644,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 447 - job_id: j7gj3rnxp + job_id: jw5601w75 job_status: Passed torchscript_onnx: - inference_time: 4523.0 - throughput: 221.09219544550078 + inference_time: 4456.0 + throughput: 224.4165170556553 estimated_peak_memory_range: - min: 79282176 - max: 79282176 + min: 80506880 + max: 80506880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -659,7 +659,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 462 - job_id: j1p8jeqx5 + job_id: jopr7y0eg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -668,4 +668,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:33:54Z' + timestamp: '2024-08-26T23:51:37Z' diff --git a/qai_hub_models/models/wideresnet50/perf.yaml b/qai_hub_models/models/wideresnet50/perf.yaml index 344aaabb..9d24a36f 100644 --- a/qai_hub_models/models/wideresnet50/perf.yaml +++ b/qai_hub_models/models/wideresnet50/perf.yaml @@ -45,11 +45,11 @@ models: - name: WideResNet50 performance_metrics: - torchscript_onnx_tflite: - inference_time: 4870.0 - throughput: 205.3388090349076 + inference_time: 4866.0 + throughput: 205.5076037813399 estimated_peak_memory_range: min: 16384 - max: 1594144 + max: 2291440 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jz57o43lg + job_id: jz5wr98jp job_status: Passed torchscript_onnx_qnn: - inference_time: 5663.0 - throughput: 176.58484901995408 + inference_time: 5669.0 + throughput: 176.3979537837361 estimated_peak_memory_range: - min: 630784 - max: 280815328 + min: 16384 + max: 311520408 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jqpy8x07g + job_id: jo5mlzdqg job_status: Passed torchscript_onnx: - inference_time: 5416.0 - throughput: 184.6381093057607 + inference_time: 5455.0 + throughput: 183.3180568285976 estimated_peak_memory_range: - min: 638976 - max: 3111928 + min: 16384 + max: 169785048 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: jwgodrqx5 + job_id: jn5qd2mmg job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:32:50Z' + timestamp: '2024-08-26T23:50:41Z' - torchscript_onnx_tflite: - inference_time: 3606.0 - throughput: 277.31558513588465 + inference_time: 3604.0 + throughput: 277.4694783573807 estimated_peak_memory_range: min: 16384 - max: 101954320 + max: 102748240 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jqp4e10vg + job_id: jmg9q4kvp job_status: Passed torchscript_onnx_qnn: - inference_time: 4154.0 - throughput: 240.73182474723157 + inference_time: 4146.0 + throughput: 241.196333815726 estimated_peak_memory_range: - min: 0 - max: 29155952 + min: 618496 + max: 28852864 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j2p0oj76p + job_id: jegnwe7mg job_status: Passed torchscript_onnx: - inference_time: 4121.0 - throughput: 242.6595486532395 + inference_time: 4110.0 + throughput: 243.30900243309003 estimated_peak_memory_range: - min: 606208 - max: 103962896 + min: 655360 + max: 106667104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: j1pv2dxjg + job_id: j1glqk1lp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:32:51Z' + timestamp: '2024-08-26T23:50:42Z' - torchscript_onnx_tflite: - inference_time: 4895.0 - throughput: 204.29009193054137 + inference_time: 4840.0 + throughput: 206.61157024793388 estimated_peak_memory_range: - min: 24576 - max: 2422000 + min: 28672 + max: 182997448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: j0px0421p + job_id: jnp1m87lp job_status: Passed torchscript_onnx_qnn: - inference_time: 4897.0 - throughput: 204.20665713702266 + inference_time: 4875.0 + throughput: 205.12820512820514 estimated_peak_memory_range: - min: 634880 - max: 2185480 + min: 626688 + max: 1873080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jogk64m25 + job_id: jep2zmvmp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:32:44Z' + timestamp: '2024-08-26T23:50:37Z' - torchscript_onnx_tflite: - inference_time: 7214.0 - throughput: 138.6193512614361 + inference_time: 7134.0 + throughput: 140.17381553125875 estimated_peak_memory_range: - min: 20480 - max: 91854320 + min: 24576 + max: 92258416 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jo5m9mywg + job_id: jvgdmv8lg job_status: Passed torchscript_onnx_qnn: - inference_time: 7250.0 - throughput: 137.93103448275863 + inference_time: 7233.0 + throughput: 138.25521913452232 estimated_peak_memory_range: - min: 622592 - max: 23208448 + min: 618496 + max: 25426352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1p3o92lp + job_id: jogkky9og job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:32:49Z' + timestamp: '2024-08-26T23:50:40Z' - torchscript_onnx_tflite: - inference_time: 4866.0 - throughput: 205.5076037813399 + inference_time: 4852.0 + throughput: 206.10057708161582 estimated_peak_memory_range: - min: 20480 - max: 2076080 + min: 40960 + max: 2529528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jegn1n8rp + job_id: jz578d7rp job_status: Passed torchscript_onnx_qnn: - inference_time: 4903.0 - throughput: 203.95676116663267 + inference_time: 4890.0 + throughput: 204.49897750511246 estimated_peak_memory_range: - min: 667648 - max: 1985872 + min: 655360 + max: 2156944 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jn5q4yo4g + job_id: jqpyyd74p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:32:46Z' + timestamp: '2024-08-26T23:50:38Z' - torchscript_onnx_tflite: - inference_time: 4882.0 - throughput: 204.83408439164276 + inference_time: 4858.0 + throughput: 205.8460271716756 estimated_peak_memory_range: min: 16384 - max: 2037456 + max: 2154904 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: joprx0j9p + job_id: jqp42w9lg job_status: Passed torchscript_onnx_qnn: - inference_time: 4907.0 - throughput: 203.79050336254332 + inference_time: 4890.0 + throughput: 204.49897750511246 estimated_peak_memory_range: - min: 634880 - max: 1947032 + min: 638976 + max: 1974600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1glwxr8p + job_id: j2p0xrvep job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:32:47Z' + timestamp: '2024-08-26T23:50:38Z' - torchscript_onnx_tflite: - inference_time: 4881.0 - throughput: 204.8760499897562 + inference_time: 4855.0 + throughput: 205.97322348094747 estimated_peak_memory_range: min: 24576 - max: 2096584 + max: 2297392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 79 - job_id: jep2own4g + job_id: j0pxz1d95 job_status: Passed torchscript_onnx_qnn: - inference_time: 4850.0 - throughput: 206.18556701030928 + inference_time: 4842.0 + throughput: 206.52622883106156 estimated_peak_memory_range: min: 634880 - max: 1874704 + max: 1915776 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: jw56o7l05 + job_id: j1p8k748p job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:32:48Z' + timestamp: '2024-08-26T23:50:39Z' - torchscript_onnx_qnn: - inference_time: 4696.0 - throughput: 212.94718909710392 + inference_time: 4685.0 + throughput: 213.4471718249733 estimated_peak_memory_range: min: 602112 max: 602112 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 126 - job_id: j1p8jxvx5 + job_id: jopr7yneg job_status: Passed torchscript_onnx: - inference_time: 5104.0 - throughput: 195.92476489028212 + inference_time: 5101.0 + throughput: 196.0399921584003 estimated_peak_memory_range: - min: 140283904 - max: 140283904 + min: 140357632 + max: 140357632 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 128 - job_id: j7gj374xp + job_id: jw5601d75 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:32:52Z' + timestamp: '2024-08-26T23:50:43Z' diff --git a/qai_hub_models/models/wideresnet50_quantized/perf.yaml b/qai_hub_models/models/wideresnet50_quantized/perf.yaml index 64152eb3..789dc819 100644 --- a/qai_hub_models/models/wideresnet50_quantized/perf.yaml +++ b/qai_hub_models/models/wideresnet50_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: WideResNet50-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 1800.0 - throughput: 555.5555555555555 + inference_time: 1773.0 + throughput: 564.0157924421884 estimated_peak_memory_range: - min: 12288 - max: 7528768 + min: 32768 + max: 623939016 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jo5m9m2qg + job_id: jnp1m83kp job_status: Passed torchscript_onnx_qnn: - inference_time: 2049.0 - throughput: 488.0429477794046 + inference_time: 2044.0 + throughput: 489.23679060665364 estimated_peak_memory_range: - min: 16384 - max: 9267976 + min: 12288 + max: 556056656 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j1glwxzlp + job_id: jegnweomg job_status: Passed torchscript_onnx: - inference_time: 2001.0 - throughput: 499.7501249375312 + inference_time: 1998.0 + throughput: 500.5005005005005 estimated_peak_memory_range: - min: 12288 - max: 85990448 + min: 32768 + max: 867359896 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: jmg9om1vg + job_id: jw5601r75 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:32:06Z' + timestamp: '2024-08-26T23:50:01Z' - torchscript_onnx_tflite: - inference_time: 1388.0 - throughput: 720.4610951008646 + inference_time: 1348.0 + throughput: 741.839762611276 estimated_peak_memory_range: min: 12288 - max: 57860656 + max: 59192848 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jegn1nymp + job_id: jvgdmv0kg job_status: Passed torchscript_onnx_qnn: - inference_time: 1546.0 - throughput: 646.8305304010349 + inference_time: 1533.0 + throughput: 652.3157208088714 estimated_peak_memory_range: min: 0 - max: 18726576 + max: 18704144 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jw56o7j75 + job_id: jopr7yoeg job_status: Passed torchscript_onnx: - inference_time: 1550.0 - throughput: 645.1612903225806 + inference_time: 1595.0 + throughput: 626.9592476489029 estimated_peak_memory_range: - min: 155648 - max: 74161888 + min: 12288 + max: 74736912 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: jnp1ojll5 + job_id: j1p3rmxzp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:32:07Z' + timestamp: '2024-08-26T23:50:01Z' - torchscript_onnx_tflite: - inference_time: 1794.0 - throughput: 557.4136008918617 + inference_time: 1779.0 + throughput: 562.1135469364812 estimated_peak_memory_range: min: 12288 - max: 24285304 + max: 17931368 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: joprx0qep + job_id: jz5wr9xjp job_status: Passed torchscript_onnx_qnn: - inference_time: 1930.0 - throughput: 518.1347150259068 + inference_time: 1885.0 + throughput: 530.5039787798408 estimated_peak_memory_range: - min: 192512 - max: 1767024 + min: 188416 + max: 1623720 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jwgodr0d5 + job_id: jqpyydq4p job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:31:59Z' + timestamp: '2024-08-26T23:49:54Z' - torchscript_onnx_tflite: - inference_time: 2189.0 - throughput: 456.82960255824577 + inference_time: 2176.0 + throughput: 459.55882352941177 estimated_peak_memory_range: - min: 16384 - max: 59019408 + min: 20480 + max: 60789184 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jep2ow6mg + job_id: jmg9q48vp job_status: Passed torchscript_onnx_qnn: - inference_time: 2476.0 - throughput: 403.8772213247173 + inference_time: 2475.0 + throughput: 404.04040404040404 estimated_peak_memory_range: - min: 188416 - max: 21660112 + min: 167936 + max: 21777008 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jygzzm96g + job_id: jn5qd2zmg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:32:04Z' + timestamp: '2024-08-26T23:49:58Z' - torchscript_onnx_tflite: - inference_time: 1798.0 - throughput: 556.1735261401557 + inference_time: 1779.0 + throughput: 562.1135469364812 estimated_peak_memory_range: - min: 16384 - max: 139286624 + min: 12288 + max: 1746968 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jqpy8xw4g + job_id: jvgdmv0lg job_status: Passed torchscript_onnx_qnn: - inference_time: 1930.0 - throughput: 518.1347150259068 + inference_time: 1880.0 + throughput: 531.9148936170212 estimated_peak_memory_range: - min: 172032 - max: 1481744 + min: 200704 + max: 1475336 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j1pv2domg + job_id: j2p0xrdep job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:32:00Z' + timestamp: '2024-08-26T23:49:55Z' - torchscript_onnx_tflite: - inference_time: 1800.0 - throughput: 555.5555555555555 + inference_time: 1772.0 + throughput: 564.3340857787811 estimated_peak_memory_range: - min: 24576 - max: 27120392 + min: 12288 + max: 1433680 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: j2p0ojqep + job_id: jz578dkrp job_status: Passed torchscript_onnx_qnn: - inference_time: 1935.0 - throughput: 516.795865633075 + inference_time: 1884.0 + throughput: 530.7855626326964 estimated_peak_memory_range: - min: 184320 - max: 1448640 + min: 180224 + max: 1412800 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j7gj37m8p + job_id: j1p8k768p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:32:02Z' + timestamp: '2024-08-26T23:49:56Z' - torchscript_onnx_tflite: - inference_time: 1804.0 - throughput: 554.3237250554324 + inference_time: 1779.0 + throughput: 562.1135469364812 estimated_peak_memory_range: - min: 28672 - max: 166005704 + min: 12288 + max: 1620048 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: j1p8jx985 + job_id: jqp42wmlg job_status: Passed torchscript_onnx_qnn: - inference_time: 1931.0 - throughput: 517.8663904712585 + inference_time: 1929.0 + throughput: 518.4033177812338 estimated_peak_memory_range: - min: 172032 - max: 1430600 + min: 196608 + max: 1732432 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jlpe6z10g + job_id: jogkkyoog job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:32:03Z' + timestamp: '2024-08-26T23:49:57Z' - torchscript_onnx_tflite: - inference_time: 7819.0 - throughput: 127.89359253101419 + inference_time: 7825.0 + throughput: 127.79552715654953 estimated_peak_memory_range: min: 12288 - max: 29496112 + max: 29848336 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jogk64no5 + job_id: j0pxz1395 job_status: Passed torchscript_onnx_qnn: - inference_time: 10098.0 - throughput: 99.02951079421668 + inference_time: 10127.0 + throughput: 98.74592673052237 estimated_peak_memory_range: - min: 167936 - max: 8317008 + min: 172032 + max: 8321232 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: jz5wy7vjg + job_id: j1glqkolp job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:32:05Z' + timestamp: '2024-08-26T23:50:00Z' - torchscript_onnx_tflite: - inference_time: 23870.0 - throughput: 41.89359028068706 + inference_time: 23959.0 + throughput: 41.73796903042698 estimated_peak_memory_range: - min: 53248 - max: 2209880 + min: 36864 + max: 3847024 primary_compute_unit: NPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 82 - job_id: jn5q4ykmg + job_id: jo5mlzoqg job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,13 +406,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:31:54Z' + timestamp: '2024-08-26T23:49:50Z' - torchscript_onnx_qnn: - inference_time: 1874.0 - throughput: 533.6179295624333 + inference_time: 1841.0 + throughput: 543.1830526887561 estimated_peak_memory_range: - min: 380928 - max: 380928 + min: 286720 + max: 286720 primary_compute_unit: NPU precision: int8 layer_info: @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 78 - job_id: j1p3o93zp + job_id: jep2zm4mp job_status: Passed torchscript_onnx: - inference_time: 1818.0 - throughput: 550.05500550055 + inference_time: 1839.0 + throughput: 543.773790103317 estimated_peak_memory_range: - min: 73474048 - max: 73474048 + min: 73555968 + max: 73555968 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 83 - job_id: jvgd639lp + job_id: jwgo9vodg job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:32:09Z' + timestamp: '2024-08-26T23:50:02Z' diff --git a/qai_hub_models/models/xlsr/perf.yaml b/qai_hub_models/models/xlsr/perf.yaml index 9094ca6a..40e1deea 100644 --- a/qai_hub_models/models/xlsr/perf.yaml +++ b/qai_hub_models/models/xlsr/perf.yaml @@ -45,11 +45,11 @@ models: - name: XLSR performance_metrics: - torchscript_onnx_tflite: - inference_time: 2510.0 - throughput: 398.40637450199205 + inference_time: 2435.0 + throughput: 410.6776180698152 estimated_peak_memory_range: - min: 24576 - max: 17452864 + min: 28672 + max: 1380352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 16 - job_id: joprx0kep + job_id: jz578dvqp job_status: Passed torchscript_onnx_qnn: - inference_time: 1363.0 - throughput: 733.6757153338225 + inference_time: 1358.0 + throughput: 736.3770250368188 estimated_peak_memory_range: - min: 12288 - max: 4891736 + min: 2109440 + max: 75239288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: j1glwxnlp + job_id: jqpyydnrp job_status: Passed torchscript_onnx: - inference_time: 1554.0 - throughput: 643.5006435006435 + inference_time: 1530.0 + throughput: 653.59477124183 estimated_peak_memory_range: - min: 212992 - max: 2384592 + min: 126976 + max: 59778952 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 23 - job_id: jz5wy7ojg + job_id: jwgo9vm4g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:29:37Z' + timestamp: '2024-08-26T23:47:34Z' - torchscript_onnx_tflite: - inference_time: 1791.0 - throughput: 558.3472920156337 + inference_time: 1692.0 + throughput: 591.016548463357 estimated_peak_memory_range: min: 0 - max: 21732416 + max: 22246528 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 16 - job_id: jep2ow8mg + job_id: jqp42wjqg job_status: Passed torchscript_onnx_qnn: - inference_time: 839.0 - throughput: 1191.8951132300358 + inference_time: 845.0 + throughput: 1183.4319526627219 estimated_peak_memory_range: min: 208896 - max: 11697952 + max: 12139408 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: jw56o7675 + job_id: j2p0xrk2p job_status: Passed torchscript_onnx: - inference_time: 999.0 - throughput: 1001.001001001001 + inference_time: 984.0 + throughput: 1016.260162601626 estimated_peak_memory_range: min: 0 - max: 23230464 + max: 22131936 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 23 - job_id: jmg9omvvg + job_id: j1pvnw47g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:29:38Z' + timestamp: '2024-08-26T23:47:34Z' - torchscript_onnx_tflite: - inference_time: 2536.0 - throughput: 394.3217665615142 + inference_time: 2490.0 + throughput: 401.60642570281124 estimated_peak_memory_range: - min: 28672 - max: 77800464 + min: 24576 + max: 1382688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 16 - job_id: jqpy8xe4g + job_id: j0pxz1ej5 job_status: Passed torchscript_onnx_qnn: - inference_time: 1351.0 - throughput: 740.1924500370096 + inference_time: 1368.0 + throughput: 730.9941520467836 estimated_peak_memory_range: - min: 229376 - max: 1490488 + min: 233472 + max: 1738672 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: jwgodryd5 + job_id: jogkkydyg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:29:31Z' + timestamp: '2024-08-26T23:47:29Z' - torchscript_onnx_tflite: - inference_time: 3378.0 - throughput: 296.0331557134399 + inference_time: 4353.0 + throughput: 229.72662531587412 estimated_peak_memory_range: - min: 12640256 - max: 36495440 + min: 12636160 + max: 36040080 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 16 - job_id: j2p0ojyep + job_id: jo5mlzvyg job_status: Passed torchscript_onnx_qnn: - inference_time: 1591.0 - throughput: 628.5355122564425 + inference_time: 1587.0 + throughput: 630.119722747322 estimated_peak_memory_range: min: 208896 - max: 13994032 + max: 13497104 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: jygzzme6g + job_id: j1p3rm8xp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:29:36Z' + timestamp: '2024-08-26T23:47:33Z' - torchscript_onnx_tflite: - inference_time: 2566.0 - throughput: 389.7116134060795 + inference_time: 2408.0 + throughput: 415.28239202657807 estimated_peak_memory_range: - min: 32768 - max: 1580880 + min: 24576 + max: 1675072 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 16 - job_id: j1p8jxo85 + job_id: jegnwexvg job_status: Passed torchscript_onnx_qnn: - inference_time: 1362.0 - throughput: 734.2143906020558 + inference_time: 1353.0 + throughput: 739.0983000739099 estimated_peak_memory_range: - min: 229376 - max: 1544832 + min: 36864 + max: 1407712 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: j1pv2d3mg + job_id: jn5qd2w7g job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:29:32Z' + timestamp: '2024-08-26T23:47:30Z' - torchscript_onnx_tflite: - inference_time: 2664.0 - throughput: 375.37537537537537 + inference_time: 2584.0 + throughput: 386.9969040247678 estimated_peak_memory_range: - min: 24576 - max: 1595336 + min: 12673024 + max: 14348032 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 16 - job_id: jogk64zo5 + job_id: jopr7y9vg job_status: Passed torchscript_onnx_qnn: - inference_time: 1348.0 - throughput: 741.839762611276 + inference_time: 1363.0 + throughput: 733.6757153338225 estimated_peak_memory_range: - min: 233472 - max: 1536104 + min: 229376 + max: 1453448 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: j7gj37x8p + job_id: j1glqk7ep job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:29:33Z' + timestamp: '2024-08-26T23:47:31Z' - torchscript_onnx_tflite: - inference_time: 2636.0 - throughput: 379.3626707132018 + inference_time: 2499.0 + throughput: 400.16006402561027 estimated_peak_memory_range: - min: 16384 - max: 1429048 + min: 28672 + max: 9314456 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 16 - job_id: jn5q4y8mg + job_id: jep2zmjxp job_status: Passed torchscript_onnx_qnn: - inference_time: 1344.0 - throughput: 744.047619047619 + inference_time: 1356.0 + throughput: 737.4631268436578 estimated_peak_memory_range: - min: 233472 - max: 1713880 + min: 229376 + max: 1872672 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: jlpe6z90g + job_id: jw5601vv5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,13 +339,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:29:35Z' + timestamp: '2024-08-26T23:47:32Z' - torchscript_onnx_qnn: - inference_time: 1487.0 - throughput: 672.4949562878278 + inference_time: 1503.0 + throughput: 665.335994677312 estimated_peak_memory_range: - min: 208896 - max: 208896 + min: 212992 + max: 212992 primary_compute_unit: NPU precision: fp16 layer_info: @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: j1p3o9kzp + job_id: j1p8k78zp job_status: Passed torchscript_onnx: - inference_time: 1537.0 - throughput: 650.6180871828237 + inference_time: 1515.0 + throughput: 660.0660066006601 estimated_peak_memory_range: - min: 8839168 - max: 8839168 + min: 8826880 + max: 8826880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 23 - job_id: jnp1oj0l5 + job_id: j7gj8l175 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:29:39Z' + timestamp: '2024-08-26T23:47:35Z' diff --git a/qai_hub_models/models/xlsr_quantized/perf.yaml b/qai_hub_models/models/xlsr_quantized/perf.yaml index 521e99b2..0173ab98 100644 --- a/qai_hub_models/models/xlsr_quantized/perf.yaml +++ b/qai_hub_models/models/xlsr_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: XLSR-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 1125.0 - throughput: 888.8888888888889 + inference_time: 1062.0 + throughput: 941.6195856873823 estimated_peak_memory_range: - min: 20480 - max: 1497448 + min: 45056 + max: 1713648 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,14 +63,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jegn1njmp + job_id: jvgdmvrkg job_status: Passed torchscript_onnx_qnn: - inference_time: 651.0 - throughput: 1536.0983102918588 + inference_time: 649.0 + throughput: 1540.8320493066255 estimated_peak_memory_range: - min: 28672 - max: 3006712 + min: 12288 + max: 3539208 primary_compute_unit: NPU precision: int8 layer_info: @@ -78,14 +78,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: jw56o7k75 + job_id: j2p0xre2p job_status: Passed torchscript_onnx: - inference_time: 744.0 - throughput: 1344.0860215053763 + inference_time: 779.0 + throughput: 1283.6970474967907 estimated_peak_memory_range: - min: 16384 - max: 13782384 + min: 69632 + max: 49881560 primary_compute_unit: NPU precision: int8 layer_info: @@ -93,7 +93,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: jnp1ojrl5 + job_id: j7gj8lk75 job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -102,13 +102,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:28:59Z' + timestamp: '2024-08-26T23:47:00Z' - torchscript_onnx_tflite: - inference_time: 913.0 - throughput: 1095.290251916758 + inference_time: 870.0 + throughput: 1149.4252873563219 estimated_peak_memory_range: min: 0 - max: 22506160 + max: 22756288 primary_compute_unit: NPU precision: int8 layer_info: @@ -116,14 +116,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: joprx0zep + job_id: jz578djqp job_status: Passed torchscript_onnx_qnn: - inference_time: 447.0 - throughput: 2237.136465324385 + inference_time: 448.0 + throughput: 2232.1428571428573 estimated_peak_memory_range: min: 65536 - max: 13032656 + max: 15680832 primary_compute_unit: NPU precision: int8 layer_info: @@ -131,14 +131,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: j1p3o9yzp + job_id: j1p8k7wzp job_status: Passed torchscript_onnx: - inference_time: 566.0 - throughput: 1766.7844522968198 + inference_time: 546.0 + throughput: 1831.5018315018315 estimated_peak_memory_range: min: 0 - max: 27306848 + max: 27252304 primary_compute_unit: NPU precision: int8 layer_info: @@ -146,7 +146,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: jvgd63jlp + job_id: jlpenv47p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -155,13 +155,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:29:01Z' + timestamp: '2024-08-26T23:47:01Z' - torchscript_onnx_tflite: - inference_time: 1199.0 - throughput: 834.0283569641368 + inference_time: 1056.0 + throughput: 946.969696969697 estimated_peak_memory_range: - min: 16384 - max: 1225872 + min: 12288 + max: 90716040 primary_compute_unit: NPU precision: int8 layer_info: @@ -169,14 +169,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jep2ow2mg + job_id: jqp42wxqg job_status: Passed torchscript_onnx_qnn: - inference_time: 508.0 - throughput: 1968.5039370078741 + inference_time: 507.0 + throughput: 1972.3865877712033 estimated_peak_memory_range: - min: 86016 - max: 1796584 + min: 81920 + max: 1354080 primary_compute_unit: NPU precision: int8 layer_info: @@ -184,7 +184,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: j1pv2djmg + job_id: jn5qd297g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -193,13 +193,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:28:52Z' + timestamp: '2024-08-26T23:46:54Z' - torchscript_onnx_tflite: - inference_time: 1659.0 - throughput: 602.7727546714889 + inference_time: 1341.0 + throughput: 745.7121551081283 estimated_peak_memory_range: - min: 20480 - max: 22637776 + min: 16384 + max: 24488944 primary_compute_unit: NPU precision: int8 layer_info: @@ -207,14 +207,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jqpy8x94g + job_id: j0pxz17j5 job_status: Passed torchscript_onnx_qnn: - inference_time: 716.0 - throughput: 1396.6480446927374 + inference_time: 710.0 + throughput: 1408.4507042253522 estimated_peak_memory_range: min: 61440 - max: 16380464 + max: 14136000 primary_compute_unit: NPU precision: int8 layer_info: @@ -222,7 +222,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: jz5wy7jjg + job_id: jwgo9ve4g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -231,13 +231,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:28:57Z' + timestamp: '2024-08-26T23:46:58Z' - torchscript_onnx_tflite: - inference_time: 1122.0 - throughput: 891.2655971479501 + inference_time: 1781.0 + throughput: 561.4823133071309 estimated_peak_memory_range: - min: 20480 - max: 3984696 + min: 28672 + max: 4746536 primary_compute_unit: NPU precision: int8 layer_info: @@ -245,14 +245,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: j2p0ojnep + job_id: jo5mlzwyg job_status: Passed torchscript_onnx_qnn: - inference_time: 514.0 - throughput: 1945.5252918287938 + inference_time: 512.0 + throughput: 1953.125 estimated_peak_memory_range: - min: 77824 - max: 1374440 + min: 81920 + max: 1572280 primary_compute_unit: NPU precision: int8 layer_info: @@ -260,7 +260,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: j7gj37j8p + job_id: j1glqkeep job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -269,13 +269,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:28:54Z' + timestamp: '2024-08-26T23:46:55Z' - torchscript_onnx_tflite: - inference_time: 1118.0 - throughput: 894.4543828264758 + inference_time: 1044.0 + throughput: 957.8544061302682 estimated_peak_memory_range: - min: 28672 - max: 1492680 + min: 24576 + max: 1620248 primary_compute_unit: NPU precision: int8 layer_info: @@ -283,14 +283,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: j1p8jxl85 + job_id: jegnwervg job_status: Passed torchscript_onnx_qnn: - inference_time: 511.0 - throughput: 1956.9471624266146 + inference_time: 510.0 + throughput: 1960.7843137254902 estimated_peak_memory_range: - min: 77824 - max: 1491368 + min: 86016 + max: 1381664 primary_compute_unit: NPU precision: int8 layer_info: @@ -298,7 +298,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: jlpe6zj0g + job_id: jw5601qv5 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -307,13 +307,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:28:55Z' + timestamp: '2024-08-26T23:46:56Z' - torchscript_onnx_tflite: - inference_time: 1112.0 - throughput: 899.2805755395683 + inference_time: 1059.0 + throughput: 944.2870632672333 estimated_peak_memory_range: - min: 32768 - max: 14833304 + min: 12288 + max: 23984312 primary_compute_unit: NPU precision: int8 layer_info: @@ -321,14 +321,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jogk64jo5 + job_id: jopr7y1vg job_status: Passed torchscript_onnx_qnn: - inference_time: 506.0 - throughput: 1976.2845849802372 + inference_time: 513.0 + throughput: 1949.317738791423 estimated_peak_memory_range: - min: 81920 - max: 1301600 + min: 90112 + max: 1314160 primary_compute_unit: NPU precision: int8 layer_info: @@ -336,7 +336,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: jygzzm16g + job_id: j1p3rmqxp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -345,13 +345,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:28:56Z' + timestamp: '2024-08-26T23:46:57Z' - torchscript_onnx_tflite: - inference_time: 2459.0 - throughput: 406.669377795852 + inference_time: 2346.0 + throughput: 426.25745950554136 estimated_peak_memory_range: - min: 3203072 - max: 19365664 + min: 12288 + max: 16249232 primary_compute_unit: NPU precision: int8 layer_info: @@ -359,14 +359,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 3 total_layers: 19 - job_id: jn5q4yjmg + job_id: jep2zm3xp job_status: Passed torchscript_onnx_qnn: - inference_time: 1095.0 - throughput: 913.2420091324201 + inference_time: 1100.0 + throughput: 909.0909090909091 estimated_peak_memory_range: min: 12288 - max: 7190768 + max: 7937296 primary_compute_unit: NPU precision: int8 layer_info: @@ -374,7 +374,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: jmg9om6vg + job_id: j1pvnwz7g job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -383,13 +383,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:28:58Z' + timestamp: '2024-08-26T23:46:59Z' - torchscript_onnx_tflite: - inference_time: 16596.0 - throughput: 60.25548324897566 + inference_time: 16392.0 + throughput: 61.00536847242557 estimated_peak_memory_range: - min: 6332416 - max: 19343664 + min: 6504448 + max: 12217680 primary_compute_unit: GPU precision: int8 layer_info: @@ -397,7 +397,7 @@ models: layers_on_gpu: 9 layers_on_cpu: 5 total_layers: 19 - job_id: j1glwxjlp + job_id: jqpyydvrp job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -406,10 +406,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:28:48Z' + timestamp: '2024-08-26T23:46:50Z' - torchscript_onnx_qnn: - inference_time: 706.0 - throughput: 1416.4305949008499 + inference_time: 632.0 + throughput: 1582.2784810126582 estimated_peak_memory_range: min: 61440 max: 61440 @@ -420,14 +420,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 16 - job_id: jwgodrjd5 + job_id: jogkkyryg job_status: Passed torchscript_onnx: - inference_time: 794.0 - throughput: 1259.4458438287154 + inference_time: 756.0 + throughput: 1322.7513227513227 estimated_peak_memory_range: - min: 3379200 - max: 3379200 + min: 3567616 + max: 3567616 primary_compute_unit: NPU precision: int8 layer_info: @@ -435,7 +435,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 21 - job_id: jz57o4zrg + job_id: jygz07vz5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -444,4 +444,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:29:02Z' + timestamp: '2024-08-26T23:47:02Z' diff --git a/qai_hub_models/models/yolonas/README.md b/qai_hub_models/models/yolonas/README.md index e35a296d..ae7002aa 100644 --- a/qai_hub_models/models/yolonas/README.md +++ b/qai_hub_models/models/yolonas/README.md @@ -46,8 +46,8 @@ script requires access to Deployment instructions for Qualcomm® AI Hub. ## License - The license for the original implementation of Yolo-NAS can be found - [here](https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.YOLONAS.md). -- The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf) + [here](https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.md). +- The license for the compiled assets for on-device deployment can be found [here](https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.md) ## References * [YOLO-NAS by Deci Achieves SOTA Performance on Object Detection Using Neural Architecture Search](https://deci.ai/blog/yolo-nas-object-detection-foundation-model/) diff --git a/qai_hub_models/models/yolonas/info.yaml b/qai_hub_models/models/yolonas/info.yaml index 64998cc1..8d36bddc 100644 --- a/qai_hub_models/models/yolonas/info.yaml +++ b/qai_hub_models/models/yolonas/info.yaml @@ -11,8 +11,8 @@ tags: - real-time research_paper: https://deci.ai/blog/yolo-nas-object-detection-foundation-model/ research_paper_title: 'YOLO-NAS by Deci Achieves SOTA Performance on Object Detection Using Neural Architecture Search' -license: https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.YOLONAS.md -deploy_license: https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf +license: https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.md +deploy_license: https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.md source_repo: https://github.com/Deci-AI/super-gradients technical_details: Model checkpoint: YoloNAS Small diff --git a/qai_hub_models/models/yolonas/perf.yaml b/qai_hub_models/models/yolonas/perf.yaml index 88b22d72..430be70b 100644 --- a/qai_hub_models/models/yolonas/perf.yaml +++ b/qai_hub_models/models/yolonas/perf.yaml @@ -45,11 +45,11 @@ models: - name: Yolo-NAS performance_metrics: - torchscript_onnx_tflite: - inference_time: 10797.0 - throughput: 92.61831990367695 + inference_time: 10850.0 + throughput: 92.16589861751152 estimated_peak_memory_range: - min: 217088 - max: 4271776 + min: 40960 + max: 3276872 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 201 - job_id: jep2ow1mg + job_id: jqp42wwqg job_status: Passed torchscript_onnx_qnn: - inference_time: 15128.0 - throughput: 66.10259122157589 + inference_time: 14890.0 + throughput: 67.15916722632639 estimated_peak_memory_range: - min: 4960256 - max: 21912600 + min: 4931584 + max: 22000880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: jw56o7x75 + job_id: j2p0xr12p job_status: Passed torchscript_onnx: - inference_time: 11787.0 - throughput: 84.83922965979468 + inference_time: 12043.0 + throughput: 83.03578842481109 estimated_peak_memory_range: - min: 24576 - max: 26711504 + min: 7692288 + max: 10596968 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 293 - job_id: jmg9omrvg + job_id: j1pvnw17g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:28:17Z' + timestamp: '2024-08-26T23:46:22Z' - torchscript_onnx_tflite: - inference_time: 7354.0 - throughput: 135.98041881968996 + inference_time: 7320.0 + throughput: 136.6120218579235 estimated_peak_memory_range: - min: 217088 - max: 101247360 + min: 245760 + max: 101206112 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 201 - job_id: jqpy8xl4g + job_id: j0pxz11j5 job_status: Passed torchscript_onnx_qnn: - inference_time: 10069.0 - throughput: 99.31472837421789 + inference_time: 10074.0 + throughput: 99.26543577526306 estimated_peak_memory_range: min: 4952064 - max: 33094176 + max: 33528880 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: j1p3o9dzp + job_id: j1p8k73zp job_status: Passed torchscript_onnx: - inference_time: 8304.0 - throughput: 120.42389210019267 + inference_time: 7969.0 + throughput: 125.48625925461162 estimated_peak_memory_range: - min: 9416704 - max: 112716288 + min: 9383936 + max: 114783376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 293 - job_id: jnp1oj9l5 + job_id: j7gj8l075 job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:28:18Z' + timestamp: '2024-08-26T23:46:23Z' - torchscript_onnx_tflite: - inference_time: 10823.0 - throughput: 92.39582370876836 + inference_time: 10745.0 + throughput: 93.06654257794322 estimated_peak_memory_range: - min: 12288 - max: 4151808 + min: 249856 + max: 4457624 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 201 - job_id: j2p0ojwep + job_id: jo5mlzzyg job_status: Passed torchscript_onnx_qnn: - inference_time: 9902.0 - throughput: 100.98969905069683 + inference_time: 10086.0 + throughput: 99.14733293674401 estimated_peak_memory_range: - min: 4988928 - max: 6549016 + min: 4993024 + max: 6710352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: j1pv2d8mg + job_id: jn5qd277g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:28:11Z' + timestamp: '2024-08-26T23:46:18Z' - torchscript_onnx_tflite: - inference_time: 13973.0 - throughput: 71.56659271452087 + inference_time: 13990.0 + throughput: 71.4796283059328 estimated_peak_memory_range: - min: 270336 - max: 98005840 + min: 253952 + max: 97519760 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 201 - job_id: j1p8jxn85 + job_id: jegnwe9vg job_status: Passed torchscript_onnx_qnn: - inference_time: 17721.0 - throughput: 56.43022402798939 + inference_time: 18137.0 + throughput: 55.13591001819485 estimated_peak_memory_range: - min: 4952064 - max: 32280240 + min: 4960256 + max: 35704544 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: jz5wy7kjg + job_id: jwgo9v14g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:28:16Z' + timestamp: '2024-08-26T23:46:22Z' - torchscript_onnx_tflite: - inference_time: 10812.0 - throughput: 92.4898261191269 + inference_time: 10751.0 + throughput: 93.01460329271696 estimated_peak_memory_range: - min: 172032 - max: 7265192 + min: 221184 + max: 2961688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 201 - job_id: jogk641o5 + job_id: jopr7y4vg job_status: Passed torchscript_onnx_qnn: - inference_time: 9857.0 - throughput: 101.45074566298062 + inference_time: 10072.0 + throughput: 99.28514694201748 estimated_peak_memory_range: - min: 4956160 - max: 6275280 + min: 4988928 + max: 6337376 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: j7gj3798p + job_id: j1glqk0ep job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:28:13Z' + timestamp: '2024-08-26T23:46:19Z' - torchscript_onnx_tflite: - inference_time: 10758.0 - throughput: 92.95408068414203 + inference_time: 10860.0 + throughput: 92.08103130755065 estimated_peak_memory_range: - min: 245760 - max: 5498848 + min: 237568 + max: 7072272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 201 - job_id: jn5q4ynmg + job_id: jep2zm7xp job_status: Passed torchscript_onnx_qnn: - inference_time: 9851.0 - throughput: 101.51253679829459 + inference_time: 10080.0 + throughput: 99.2063492063492 estimated_peak_memory_range: - min: 4980736 - max: 6631696 + min: 4988928 + max: 6145136 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: jlpe6zq0g + job_id: jw56013v5 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:28:14Z' + timestamp: '2024-08-26T23:46:20Z' - torchscript_onnx_tflite: - inference_time: 10730.0 - throughput: 93.19664492078286 + inference_time: 10673.0 + throughput: 93.694368968425 estimated_peak_memory_range: - min: 241664 - max: 4147792 + min: 12288 + max: 13396112 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 201 - job_id: j1glwxdlp + job_id: jqpyyd4rp job_status: Passed torchscript_onnx_qnn: - inference_time: 9910.0 - throughput: 100.90817356205852 + inference_time: 10001.0 + throughput: 99.9900009999 estimated_peak_memory_range: - min: 4997120 - max: 6449704 + min: 4984832 + max: 6274096 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: jygzzm66g + job_id: j1p3rm4xp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:28:15Z' + timestamp: '2024-08-26T23:46:21Z' - torchscript_onnx_qnn: - inference_time: 10541.0 - throughput: 94.86765961483731 + inference_time: 11070.0 + throughput: 90.3342366757001 estimated_peak_memory_range: min: 4923392 max: 4923392 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 289 - job_id: jwgodrxd5 + job_id: jogkkylyg job_status: Passed torchscript_onnx: - inference_time: 12167.0 - throughput: 82.18952905399853 + inference_time: 12098.0 + throughput: 82.65829062654984 estimated_peak_memory_range: - min: 23289856 - max: 23289856 + min: 23498752 + max: 23498752 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 293 - job_id: jvgd63klp + job_id: jlpenvr7p job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:28:20Z' + timestamp: '2024-08-26T23:46:24Z' diff --git a/qai_hub_models/models/yolonas_quantized/README.md b/qai_hub_models/models/yolonas_quantized/README.md index 77c14dcf..e4686190 100644 --- a/qai_hub_models/models/yolonas_quantized/README.md +++ b/qai_hub_models/models/yolonas_quantized/README.md @@ -46,8 +46,8 @@ script requires access to Deployment instructions for Qualcomm® AI Hub. ## License - The license for the original implementation of Yolo-NAS-Quantized can be found - [here](https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.YOLONAS.md). -- The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf) + [here](https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.md). +- The license for the compiled assets for on-device deployment can be found [here](https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.md) ## References * [YOLO-NAS by Deci Achieves SOTA Performance on Object Detection Using Neural Architecture Search](https://deci.ai/blog/yolo-nas-object-detection-foundation-model/) diff --git a/qai_hub_models/models/yolonas_quantized/info.yaml b/qai_hub_models/models/yolonas_quantized/info.yaml index fc24da8a..d3102822 100644 --- a/qai_hub_models/models/yolonas_quantized/info.yaml +++ b/qai_hub_models/models/yolonas_quantized/info.yaml @@ -13,8 +13,8 @@ tags: - quantized research_paper: https://deci.ai/blog/yolo-nas-object-detection-foundation-model/ research_paper_title: 'YOLO-NAS by Deci Achieves SOTA Performance on Object Detection Using Neural Architecture Search' -license: https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.YOLONAS.md -deploy_license: https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf +license: https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.md +deploy_license: https://github.com/Deci-AI/super-gradients/blob/master/LICENSE.md source_repo: https://github.com/Deci-AI/super-gradients technical_details: Model checkpoint: YoloNAS Small diff --git a/qai_hub_models/models/yolonas_quantized/perf.yaml b/qai_hub_models/models/yolonas_quantized/perf.yaml index 84883ea4..a62410f2 100644 --- a/qai_hub_models/models/yolonas_quantized/perf.yaml +++ b/qai_hub_models/models/yolonas_quantized/perf.yaml @@ -48,11 +48,11 @@ models: - name: Yolo-NAS-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 4738.0 - throughput: 211.05951878429718 + inference_time: 4770.0 + throughput: 209.64360587002096 estimated_peak_memory_range: - min: 122880 - max: 33356416 + min: 28672 + max: 202582472 primary_compute_unit: NPU precision: int8 layer_info: @@ -60,7 +60,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 204 - job_id: j7gj3767p + job_id: j2p0xrr2p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -69,13 +69,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:27:18Z' + timestamp: '2024-08-26T23:45:30Z' - torchscript_onnx_tflite: - inference_time: 3071.0 - throughput: 325.626831650928 + inference_time: 3062.0 + throughput: 326.5839320705421 estimated_peak_memory_range: - min: 12288 - max: 75823664 + min: 16384 + max: 78184784 primary_compute_unit: NPU precision: int8 layer_info: @@ -83,7 +83,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 204 - job_id: jlpe6z07g + job_id: j1p8k77zp job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -92,13 +92,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:27:19Z' + timestamp: '2024-08-26T23:45:31Z' - torchscript_onnx_tflite: - inference_time: 4735.0 - throughput: 211.1932418162619 + inference_time: 4741.0 + throughput: 210.9259649862898 estimated_peak_memory_range: - min: 126976 - max: 10028544 + min: 122880 + max: 13942952 primary_compute_unit: NPU precision: int8 layer_info: @@ -106,7 +106,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 204 - job_id: jygzzmqzg + job_id: jogkkyyyg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -115,13 +115,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:27:20Z' + timestamp: '2024-08-26T23:45:32Z' - torchscript_onnx_tflite: - inference_time: 5167.0 - throughput: 193.53590090961873 + inference_time: 5202.0 + throughput: 192.23375624759709 estimated_peak_memory_range: - min: 155648 - max: 79234048 + min: 167936 + max: 80769232 primary_compute_unit: NPU precision: int8 layer_info: @@ -129,7 +129,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 204 - job_id: jz5wy70zg + job_id: jn5qd227g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -138,13 +138,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:27:21Z' + timestamp: '2024-08-26T23:45:33Z' - torchscript_onnx_tflite: - inference_time: 4737.0 - throughput: 211.10407430863415 + inference_time: 4809.0 + throughput: 207.94343938448742 estimated_peak_memory_range: - min: 73728 - max: 1619608 + min: 45056 + max: 20813616 primary_compute_unit: NPU precision: int8 layer_info: @@ -152,7 +152,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 204 - job_id: jmg9om7qg + job_id: j1glqkkep job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -161,13 +161,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:27:22Z' + timestamp: '2024-08-26T23:45:34Z' - torchscript_onnx_tflite: - inference_time: 4728.0 - throughput: 211.50592216582064 + inference_time: 4688.0 + throughput: 213.31058020477815 estimated_peak_memory_range: - min: 28672 - max: 210752968 + min: 106496 + max: 14998960 primary_compute_unit: NPU precision: int8 layer_info: @@ -175,7 +175,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 204 - job_id: jnp1ojkk5 + job_id: jw56011v5 job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -184,13 +184,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:27:23Z' + timestamp: '2024-08-26T23:45:35Z' - torchscript_onnx_tflite: - inference_time: 4747.0 - throughput: 210.6593638087213 + inference_time: 4791.0 + throughput: 208.7246921310791 estimated_peak_memory_range: - min: 167936 - max: 9653304 + min: 110592 + max: 18788616 primary_compute_unit: NPU precision: int8 layer_info: @@ -198,7 +198,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 204 - job_id: jvgd63ykp + job_id: j1p3rmmxp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -207,13 +207,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:27:25Z' + timestamp: '2024-08-26T23:45:36Z' - torchscript_onnx_tflite: - inference_time: 14083.0 - throughput: 71.00759781296598 + inference_time: 13743.0 + throughput: 72.76431637924762 estimated_peak_memory_range: - min: 143360 - max: 63332944 + min: 94208 + max: 69635232 primary_compute_unit: NPU precision: int8 layer_info: @@ -221,7 +221,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 204 - job_id: jz5wy70jg + job_id: jwgo9vv4g job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -230,4 +230,4 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:27:26Z' + timestamp: '2024-08-26T23:45:37Z' diff --git a/qai_hub_models/models/yolov6/perf.yaml b/qai_hub_models/models/yolov6/perf.yaml index 38e1f1ad..670deabf 100644 --- a/qai_hub_models/models/yolov6/perf.yaml +++ b/qai_hub_models/models/yolov6/perf.yaml @@ -45,11 +45,11 @@ models: - name: Yolo-v6 performance_metrics: - torchscript_onnx_tflite: - inference_time: 6140.0 - throughput: 162.86644951140065 + inference_time: 5871.0 + throughput: 170.32873445750297 estimated_peak_memory_range: - min: 40960 - max: 3868880 + min: 253952 + max: 3584784 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 182 - job_id: jygzzm0zg + job_id: jw5601nn5 job_status: Passed torchscript_onnx_qnn: - inference_time: 5387.0 - throughput: 185.63207722294413 + inference_time: 5394.0 + throughput: 185.39117538005192 estimated_peak_memory_range: - min: 6885376 - max: 19234728 + min: 4972544 + max: 16585808 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 228 - job_id: j0px04zjp + job_id: jz5wr9q4p job_status: Passed torchscript_onnx: - inference_time: 8921.0 - throughput: 112.09505660800359 + inference_time: 9146.0 + throughput: 109.33741526350317 estimated_peak_memory_range: - min: 7671808 - max: 10285232 + min: 2777088 + max: 13110384 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 231 - job_id: jogk64wy5 + job_id: jz578dxqp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:26:15Z' + timestamp: '2024-08-26T23:44:27Z' - torchscript_onnx_tflite: - inference_time: 4391.0 - throughput: 227.7385561375541 + inference_time: 4188.0 + throughput: 238.7774594078319 estimated_peak_memory_range: - min: 53248 - max: 84240240 + min: 237568 + max: 85464368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 182 - job_id: jz5wy7rzg + job_id: j1p3rmemp job_status: Passed torchscript_onnx_qnn: - inference_time: 3837.0 - throughput: 260.62027625749283 + inference_time: 3858.0 + throughput: 259.2016588906169 estimated_peak_memory_range: - min: 5029888 - max: 51720432 + min: 4931584 + max: 49920256 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 228 - job_id: jo5m9mlyg + job_id: jmg9q4wmp job_status: Passed torchscript_onnx: - inference_time: 6552.0 - throughput: 152.62515262515262 + inference_time: 6529.0 + throughput: 153.1628120692296 estimated_peak_memory_range: - min: 10412032 - max: 101860304 + min: 10682368 + max: 102927920 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 231 - job_id: jn5q4yx7g + job_id: jqp42wvqg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:26:16Z' + timestamp: '2024-08-26T23:44:28Z' - torchscript_onnx_tflite: - inference_time: 6303.0 - throughput: 158.654608916389 + inference_time: 5843.0 + throughput: 171.14495978093444 estimated_peak_memory_range: - min: 0 - max: 3460360 + min: 262144 + max: 4385736 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 182 - job_id: jmg9omqqg + job_id: jwgo9v31g job_status: Passed torchscript_onnx_qnn: - inference_time: 5051.0 - throughput: 197.98059790140567 + inference_time: 4985.0 + throughput: 200.60180541624874 estimated_peak_memory_range: - min: 4956160 - max: 6176520 + min: 4968448 + max: 6180160 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 228 - job_id: joprx0mvp + job_id: jvgdmvo6g job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:26:09Z' + timestamp: '2024-08-26T23:44:22Z' - torchscript_onnx_tflite: - inference_time: 7913.0 - throughput: 126.37432073802603 + inference_time: 7204.0 + throughput: 138.811771238201 estimated_peak_memory_range: - min: 237568 - max: 67568432 + min: 217088 + max: 70839600 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 182 - job_id: jnp1ojmk5 + job_id: j1pvnwvzg job_status: Passed torchscript_onnx_qnn: - inference_time: 6512.0 - throughput: 153.56265356265357 + inference_time: 6700.0 + throughput: 149.2537313432836 estimated_peak_memory_range: - min: 4931584 - max: 46901600 + min: 4947968 + max: 43660352 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 228 - job_id: j1p8jxdz5 + job_id: jvgdmvokg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:26:14Z' + timestamp: '2024-08-26T23:44:26Z' - torchscript_onnx_tflite: - inference_time: 6182.0 - throughput: 161.75994823681657 + inference_time: 5902.0 + throughput: 169.43409013893594 estimated_peak_memory_range: - min: 270336 - max: 3576504 + min: 249856 + max: 3893288 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 182 - job_id: jvgd63mkp + job_id: j7gj8le15 job_status: Passed torchscript_onnx_qnn: - inference_time: 5025.0 - throughput: 199.00497512437812 + inference_time: 4986.0 + throughput: 200.56157240272765 estimated_peak_memory_range: - min: 5009408 - max: 6507136 + min: 4960256 + max: 6556488 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 228 - job_id: jep2owqxg + job_id: jz5wr9qzp job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:26:10Z' + timestamp: '2024-08-26T23:44:23Z' - torchscript_onnx_tflite: - inference_time: 6118.0 - throughput: 163.45210853220007 + inference_time: 5998.0 + throughput: 166.72224074691565 estimated_peak_memory_range: - min: 262144 - max: 3903672 + min: 16384 + max: 56770264 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 182 - job_id: jz57o48qg + job_id: jlpenvk8p job_status: Passed torchscript_onnx_qnn: - inference_time: 5041.0 - throughput: 198.37333862328904 + inference_time: 5046.0 + throughput: 198.17677368212446 estimated_peak_memory_range: - min: 4960256 - max: 6222768 + min: 4993024 + max: 6178272 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 228 - job_id: jqpy8xkrg + job_id: jmg9q4wqp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:26:11Z' + timestamp: '2024-08-26T23:44:24Z' - torchscript_onnx_tflite: - inference_time: 6129.0 - throughput: 163.15875346712352 + inference_time: 5997.0 + throughput: 166.75004168751042 estimated_peak_memory_range: - min: 217088 - max: 3757240 + min: 262144 + max: 4337792 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 182 - job_id: jqp4e12qg + job_id: jygz07r45 job_status: Passed torchscript_onnx_qnn: - inference_time: 5100.0 - throughput: 196.07843137254903 + inference_time: 5117.0 + throughput: 195.42700801250732 estimated_peak_memory_range: - min: 4956160 - max: 6216392 + min: 4997120 + max: 6565624 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 228 - job_id: j2p0oj82p + job_id: jnp1m8ekp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:26:12Z' + timestamp: '2024-08-26T23:44:25Z' - torchscript_onnx_qnn: - inference_time: 5547.0 - throughput: 180.27762754642148 + inference_time: 5293.0 + throughput: 188.9287738522577 estimated_peak_memory_range: min: 4923392 max: 4923392 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 228 - job_id: jegn1ndvp + job_id: jnp1m8enp job_status: Passed torchscript_onnx: - inference_time: 9060.0 - throughput: 110.37527593818984 + inference_time: 9107.0 + throughput: 109.80564401010211 estimated_peak_memory_range: - min: 7954432 - max: 7954432 + min: 7794688 + max: 7794688 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 231 - job_id: j1glwx9ep + job_id: j0pxz1yj5 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:26:17Z' + timestamp: '2024-08-26T23:44:29Z' diff --git a/qai_hub_models/models/yolov7/export.py b/qai_hub_models/models/yolov7/export.py index 28561f11..41eb401e 100644 --- a/qai_hub_models/models/yolov7/export.py +++ b/qai_hub_models/models/yolov7/export.py @@ -216,10 +216,7 @@ def export_model( def main(): warnings.filterwarnings("ignore") parser = export_parser( - model_cls=Model, - supports_qnn=False, - supports_onnx=False, - supports_precompiled_qnn_onnx=False, + model_cls=Model, supports_qnn=False, supports_precompiled_qnn_onnx=False ) args = parser.parse_args() export_model(**vars(args)) diff --git a/qai_hub_models/models/yolov7/perf.yaml b/qai_hub_models/models/yolov7/perf.yaml index f512bde4..d73fa349 100644 --- a/qai_hub_models/models/yolov7/perf.yaml +++ b/qai_hub_models/models/yolov7/perf.yaml @@ -17,6 +17,7 @@ aggregated: - Samsung Galaxy S21 - Samsung Galaxy S21 Ultra - Samsung Galaxy S21+ + - Snapdragon X Elite CRD - QCS8550 (Proxy) - SA8775 (Proxy) - SA8650 (Proxy) @@ -34,6 +35,7 @@ aggregated: - Snapdragon® 8 Gen 2 - Snapdragon® 8 Gen 1 - Snapdragon® 888 + - Snapdragon® X Elite - Qcs8550 - Sa8775p - Sa8650p @@ -43,19 +45,34 @@ models: - name: Yolo-v7 performance_metrics: - torchscript_onnx_tflite: - inference_time: 24674.0 - throughput: 40.52849152954527 + inference_time: 17092.0 + throughput: 58.50690381465013 estimated_peak_memory_range: - min: 40108032 - max: 72866936 - primary_compute_unit: GPU + min: 49152 + max: 2603488 + primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 0 - layers_on_gpu: 145 - layers_on_cpu: 70 + layers_on_npu: 215 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 215 - job_id: jegn1nwvp + job_id: jwgo9vn1g + job_status: Passed + torchscript_onnx: + inference_time: 17844.0 + throughput: 56.04124635731899 + estimated_peak_memory_range: + min: 26329088 + max: 103491928 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 224 + layers_on_gpu: 0 + layers_on_cpu: 1 + total_layers: 225 + job_id: jep2zm96p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -64,21 +81,36 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:25:24Z' + timestamp: '2024-08-26T23:43:51Z' - torchscript_onnx_tflite: - inference_time: 18339.0 - throughput: 54.52860025083156 + inference_time: 11522.0 + throughput: 86.79048776254123 estimated_peak_memory_range: - min: 41955328 - max: 103756848 - primary_compute_unit: GPU + min: 278528 + max: 90673088 + primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 0 - layers_on_gpu: 145 - layers_on_cpu: 70 + layers_on_npu: 215 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 215 - job_id: joprx07vp + job_id: j1pvnwrzg + job_status: Passed + torchscript_onnx: + inference_time: 12401.0 + throughput: 80.638658172728 + estimated_peak_memory_range: + min: 15208448 + max: 116853600 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 224 + layers_on_gpu: 0 + layers_on_cpu: 1 + total_layers: 225 + job_id: jqpyydj0p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -87,21 +119,21 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:25:25Z' + timestamp: '2024-08-26T23:43:52Z' - torchscript_onnx_tflite: - inference_time: 24240.0 - throughput: 41.254125412541256 + inference_time: 17108.0 + throughput: 58.45218611176058 estimated_peak_memory_range: - min: 28708864 - max: 72950608 - primary_compute_unit: GPU + min: 622592 + max: 2751312 + primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 0 - layers_on_gpu: 145 - layers_on_cpu: 70 + layers_on_npu: 215 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 215 - job_id: jep2owzxg + job_id: j7gj8l215 job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -110,21 +142,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:25:26Z' + timestamp: '2024-08-26T23:43:38Z' - torchscript_onnx_tflite: - inference_time: 36471.0 - throughput: 27.41904526884374 + inference_time: 19573.0 + throughput: 51.090788330863944 estimated_peak_memory_range: - min: 229376 - max: 69799424 - primary_compute_unit: GPU + min: 671744 + max: 92745392 + primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 0 - layers_on_gpu: 145 - layers_on_cpu: 70 + layers_on_npu: 215 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 215 - job_id: jqpy8xyrg + job_id: jlpenvw8p job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -133,21 +165,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:25:27Z' + timestamp: '2024-08-26T23:43:39Z' - torchscript_onnx_tflite: - inference_time: 24655.0 - throughput: 40.55972419387548 + inference_time: 17147.0 + throughput: 58.3192395171167 estimated_peak_memory_range: - min: 27365376 - max: 72952608 - primary_compute_unit: GPU + min: 655360 + max: 3115768 + primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 0 - layers_on_gpu: 145 - layers_on_cpu: 70 + layers_on_npu: 215 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 215 - job_id: j2p0ojx2p + job_id: jygz07j45 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -156,21 +188,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:25:29Z' + timestamp: '2024-08-26T23:43:40Z' - torchscript_onnx_tflite: - inference_time: 24096.0 - throughput: 41.50066401062417 + inference_time: 17103.0 + throughput: 58.46927439630474 estimated_peak_memory_range: - min: 36454400 - max: 134118096 - primary_compute_unit: GPU + min: 667648 + max: 6723072 + primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 0 - layers_on_gpu: 145 - layers_on_cpu: 70 + layers_on_npu: 215 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 215 - job_id: j1p8jxkz5 + job_id: jz5wr934p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -179,21 +211,21 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:25:30Z' + timestamp: '2024-08-26T23:43:41Z' - torchscript_onnx_tflite: - inference_time: 24575.0 - throughput: 40.691759918616484 + inference_time: 17177.0 + throughput: 58.21738371077604 estimated_peak_memory_range: - min: 31543296 - max: 76581296 - primary_compute_unit: GPU + min: 12288 + max: 2986704 + primary_compute_unit: NPU precision: fp16 layer_info: - layers_on_npu: 0 - layers_on_gpu: 145 - layers_on_cpu: 70 + layers_on_npu: 215 + layers_on_gpu: 0 + layers_on_cpu: 0 total_layers: 215 - job_id: jogk64ky5 + job_id: jmg9q4ymp job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -202,4 +234,27 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:25:31Z' + timestamp: '2024-08-26T23:43:42Z' + - torchscript_onnx: + inference_time: 17605.0 + throughput: 56.80204487361545 + estimated_peak_memory_range: + min: 18595840 + max: 18595840 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 224 + layers_on_gpu: 0 + layers_on_cpu: 1 + total_layers: 225 + job_id: j2p0xrl0p + job_status: Passed + reference_device_info: + name: Snapdragon X Elite CRD + os: '11' + form_factor: Compute + os_name: Windows + manufacturer: Qualcomm + chipset: Snapdragon® X Elite + timestamp: '2024-08-26T23:43:53Z' diff --git a/qai_hub_models/models/yolov7_quantized/perf.yaml b/qai_hub_models/models/yolov7_quantized/perf.yaml index 61d1574f..d5a5ab66 100644 --- a/qai_hub_models/models/yolov7_quantized/perf.yaml +++ b/qai_hub_models/models/yolov7_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: Yolo-v7-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 4427.0 - throughput: 225.886604924328 + inference_time: 4395.0 + throughput: 227.53128555176337 estimated_peak_memory_range: - min: 45056 - max: 7198224 + min: 294912 + max: 1628592 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,7 +63,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 229 - job_id: jn5q4y17g + job_id: jnp1m8ynp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -72,13 +72,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:24:47Z' + timestamp: '2024-08-26T23:43:05Z' - torchscript_onnx_tflite: - inference_time: 2820.0 - throughput: 354.6099290780142 + inference_time: 2808.0 + throughput: 356.1253561253561 estimated_peak_memory_range: min: 12288 - max: 64801904 + max: 66076752 primary_compute_unit: NPU precision: int8 layer_info: @@ -86,7 +86,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 229 - job_id: j1glwx8ep + job_id: jvgdmve6g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -95,13 +95,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:24:48Z' + timestamp: '2024-08-26T23:43:06Z' - torchscript_onnx_tflite: - inference_time: 4408.0 - throughput: 226.86025408348456 + inference_time: 4355.0 + throughput: 229.6211251435132 estimated_peak_memory_range: - min: 282624 - max: 1931344 + min: 262144 + max: 1957936 primary_compute_unit: NPU precision: int8 layer_info: @@ -109,7 +109,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 229 - job_id: jw56o7mv5 + job_id: jz578d0np job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -118,13 +118,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:24:49Z' + timestamp: '2024-08-26T23:43:07Z' - torchscript_onnx_tflite: - inference_time: 4996.0 - throughput: 200.160128102482 + inference_time: 4970.0 + throughput: 201.2072434607646 estimated_peak_memory_range: - min: 335872 - max: 68658512 + min: 319488 + max: 68181200 primary_compute_unit: NPU precision: int8 layer_info: @@ -132,7 +132,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 229 - job_id: j1p3o97xp + job_id: jqp42wk2g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -141,13 +141,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:24:50Z' + timestamp: '2024-08-26T23:43:08Z' - torchscript_onnx_tflite: - inference_time: 4411.0 - throughput: 226.70596236681024 + inference_time: 4397.0 + throughput: 227.42779167614282 estimated_peak_memory_range: - min: 282624 - max: 5330144 + min: 372736 + max: 1775272 primary_compute_unit: NPU precision: int8 layer_info: @@ -155,7 +155,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 229 - job_id: jwgodrw45 + job_id: j0pxz1n85 job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -164,13 +164,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:24:52Z' + timestamp: '2024-08-26T23:43:08Z' - torchscript_onnx_tflite: - inference_time: 4428.0 - throughput: 225.83559168925024 + inference_time: 4386.0 + throughput: 227.99817601459188 estimated_peak_memory_range: - min: 57344 - max: 1657448 + min: 282624 + max: 2089728 primary_compute_unit: NPU precision: int8 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 229 - job_id: j1pv2dm7g + job_id: jo5mlzq7g job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:24:53Z' + timestamp: '2024-08-26T23:43:09Z' - torchscript_onnx_tflite: - inference_time: 4390.0 - throughput: 227.79043280182233 + inference_time: 4382.0 + throughput: 228.20629849383843 estimated_peak_memory_range: - min: 290816 - max: 111889872 + min: 307200 + max: 4250576 primary_compute_unit: NPU precision: int8 layer_info: @@ -201,7 +201,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 229 - job_id: j7gj37y7p + job_id: jegnweljg job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -210,13 +210,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:24:54Z' + timestamp: '2024-08-26T23:43:10Z' - torchscript_onnx_tflite: - inference_time: 10200.0 - throughput: 98.03921568627452 + inference_time: 10022.0 + throughput: 99.78048293753741 estimated_peak_memory_range: min: 262144 - max: 71736192 + max: 72523984 primary_compute_unit: NPU precision: int8 layer_info: @@ -224,7 +224,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 229 - job_id: jlpe6zx7g + job_id: jopr7y8kg job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -233,13 +233,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:24:55Z' + timestamp: '2024-08-26T23:43:11Z' - torchscript_onnx_tflite: - inference_time: 97822.0 - throughput: 10.222649301793053 + inference_time: 96697.0 + throughput: 10.3415824689494 estimated_peak_memory_range: - min: 3928064 - max: 40312024 + min: 9109504 + max: 28528880 primary_compute_unit: GPU precision: int8 layer_info: @@ -247,7 +247,7 @@ models: layers_on_gpu: 127 layers_on_cpu: 70 total_layers: 229 - job_id: jygzzmyzg + job_id: jep2zm06p job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -256,4 +256,4 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:24:56Z' + timestamp: '2024-08-26T23:43:12Z' diff --git a/qai_hub_models/models/yolov8_det/perf.yaml b/qai_hub_models/models/yolov8_det/perf.yaml index 160091e0..2a0115fb 100644 --- a/qai_hub_models/models/yolov8_det/perf.yaml +++ b/qai_hub_models/models/yolov8_det/perf.yaml @@ -45,11 +45,11 @@ models: - name: YOLOv8-Detection performance_metrics: - torchscript_onnx_tflite: - inference_time: 5212.0 - throughput: 191.8649270913277 + inference_time: 5267.0 + throughput: 189.8614011771407 estimated_peak_memory_range: - min: 24576 - max: 2357680 + min: 241664 + max: 3022728 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jw56o7ov5 + job_id: jz578d2np job_status: Passed torchscript_onnx_qnn: - inference_time: 5206.0 - throughput: 192.0860545524395 + inference_time: 5292.0 + throughput: 188.96447467876038 estimated_peak_memory_range: - min: 3805184 - max: 18657920 + min: 4943872 + max: 19287512 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,14 +72,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 285 - job_id: jz5wy7yzg + job_id: jqpyydz0p job_status: Passed torchscript_onnx: - inference_time: 8466.0 - throughput: 118.11953697141507 + inference_time: 8247.0 + throughput: 121.25621438098703 estimated_peak_memory_range: - min: 3489792 - max: 9254600 + min: 9621504 + max: 12468680 primary_compute_unit: NPU precision: fp16 layer_info: @@ -87,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 289 - job_id: jegn1nqvp + job_id: jwgo9vk1g job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -96,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:24:15Z' + timestamp: '2024-08-26T23:42:34Z' - torchscript_onnx_tflite: - inference_time: 3720.0 - throughput: 268.81720430107526 + inference_time: 3699.0 + throughput: 270.3433360367667 estimated_peak_memory_range: min: 12288 - max: 82921232 + max: 86969264 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,14 +110,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: j1p3o9oxp + job_id: jqp42wn2g job_status: Passed torchscript_onnx_qnn: inference_time: 3681.0 throughput: 271.66530834012497 estimated_peak_memory_range: min: 4931584 - max: 49876512 + max: 50712720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -125,14 +125,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 285 - job_id: jmg9omoqg + job_id: j2p0xr30p job_status: Passed torchscript_onnx: - inference_time: 6121.0 - throughput: 163.371998039536 + inference_time: 6107.0 + throughput: 163.7465203864418 estimated_peak_memory_range: - min: 9445376 - max: 113620656 + min: 8568832 + max: 113027344 primary_compute_unit: NPU precision: fp16 layer_info: @@ -140,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 289 - job_id: joprx0dvp + job_id: j1pvnw0zg job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -149,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:24:16Z' + timestamp: '2024-08-26T23:42:35Z' - torchscript_onnx_tflite: - inference_time: 5221.0 - throughput: 191.5341888527102 + inference_time: 5154.0 + throughput: 194.02405898331392 estimated_peak_memory_range: - min: 278528 - max: 3965360 + min: 229376 + max: 2340624 primary_compute_unit: NPU precision: fp16 layer_info: @@ -163,14 +163,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jwgodrd45 + job_id: j0pxz1985 job_status: Passed torchscript_onnx_qnn: - inference_time: 5021.0 - throughput: 199.16351324437363 + inference_time: 5005.0 + throughput: 199.8001998001998 estimated_peak_memory_range: - min: 4956160 - max: 6113480 + min: 4960256 + max: 6660336 primary_compute_unit: NPU precision: fp16 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 285 - job_id: jvgd636kp + job_id: jogkky7vg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:24:08Z' + timestamp: '2024-08-26T23:42:29Z' - torchscript_onnx_tflite: - inference_time: 8300.0 - throughput: 120.48192771084338 + inference_time: 8543.0 + throughput: 117.05489874751258 estimated_peak_memory_range: - min: 217088 - max: 79862528 + min: 249856 + max: 79112560 primary_compute_unit: NPU precision: fp16 layer_info: @@ -201,14 +201,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: j1pv2d27g + job_id: jo5mlze7g job_status: Passed torchscript_onnx_qnn: - inference_time: 7527.0 - throughput: 132.85505513484787 + inference_time: 7764.0 + throughput: 128.7995878413189 estimated_peak_memory_range: - min: 4952064 - max: 37761808 + min: 4931584 + max: 37320368 primary_compute_unit: NPU precision: fp16 layer_info: @@ -216,7 +216,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 285 - job_id: jo5m9m9yg + job_id: j1p3rmvmp job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -225,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:24:13Z' + timestamp: '2024-08-26T23:42:33Z' - torchscript_onnx_tflite: - inference_time: 5217.0 - throughput: 191.68104274487254 + inference_time: 5214.0 + throughput: 191.79133103183736 estimated_peak_memory_range: - min: 45056 - max: 166878216 + min: 225280 + max: 6079744 primary_compute_unit: NPU precision: fp16 layer_info: @@ -239,14 +239,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: j7gj3737p + job_id: jegnwe0jg job_status: Passed torchscript_onnx_qnn: - inference_time: 5017.0 - throughput: 199.32230416583616 + inference_time: 5018.0 + throughput: 199.2825827022718 estimated_peak_memory_range: - min: 4952064 - max: 6543040 + min: 5013504 + max: 6606392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -254,7 +254,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 285 - job_id: jz57o4oqg + job_id: jn5qd2eeg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -263,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:24:09Z' + timestamp: '2024-08-26T23:42:30Z' - torchscript_onnx_tflite: - inference_time: 5208.0 - throughput: 192.01228878648234 + inference_time: 5258.0 + throughput: 190.1863826550019 estimated_peak_memory_range: - min: 225280 - max: 11015768 + min: 241664 + max: 4049504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -277,14 +277,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jlpe6z67g + job_id: jopr7y6kg job_status: Passed torchscript_onnx_qnn: - inference_time: 5050.0 - throughput: 198.01980198019803 + inference_time: 4989.0 + throughput: 200.44097013429544 estimated_peak_memory_range: - min: 4972544 - max: 6274736 + min: 4960256 + max: 6193624 primary_compute_unit: NPU precision: fp16 layer_info: @@ -292,7 +292,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 285 - job_id: jqp4e1eqg + job_id: j1glqk62p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -301,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:24:11Z' + timestamp: '2024-08-26T23:42:31Z' - torchscript_onnx_tflite: - inference_time: 5193.0 - throughput: 192.56691700365877 + inference_time: 5154.0 + throughput: 194.02405898331392 estimated_peak_memory_range: - min: 278528 - max: 2298224 + min: 36864 + max: 139087144 primary_compute_unit: NPU precision: fp16 layer_info: @@ -315,14 +315,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 290 - job_id: jygzzmzzg + job_id: jep2zmx6p job_status: Passed torchscript_onnx_qnn: - inference_time: 5059.0 - throughput: 197.667523225934 + inference_time: 5062.0 + throughput: 197.55037534571315 estimated_peak_memory_range: - min: 4960256 - max: 6222264 + min: 4964352 + max: 6684312 primary_compute_unit: NPU precision: fp16 layer_info: @@ -330,7 +330,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 285 - job_id: j0px040jp + job_id: jw5601en5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -339,10 +339,10 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:24:12Z' + timestamp: '2024-08-26T23:42:32Z' - torchscript_onnx_qnn: - inference_time: 5785.0 - throughput: 172.8608470181504 + inference_time: 5512.0 + throughput: 181.42235123367198 estimated_peak_memory_range: min: 4923392 max: 4923392 @@ -353,14 +353,14 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 285 - job_id: jnp1ojok5 + job_id: j1p8k70qp job_status: Passed torchscript_onnx: - inference_time: 8486.0 - throughput: 117.84115012962526 + inference_time: 8454.0 + throughput: 118.28720132481665 estimated_peak_memory_range: - min: 5545984 - max: 5545984 + min: 5595136 + max: 5595136 primary_compute_unit: NPU precision: fp16 layer_info: @@ -368,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 289 - job_id: jep2owdxg + job_id: j7gj8lz15 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -377,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:24:17Z' + timestamp: '2024-08-26T23:42:36Z' diff --git a/qai_hub_models/models/yolov8_det_quantized/perf.yaml b/qai_hub_models/models/yolov8_det_quantized/perf.yaml index a8b0557c..d0e10c73 100644 --- a/qai_hub_models/models/yolov8_det_quantized/perf.yaml +++ b/qai_hub_models/models/yolov8_det_quantized/perf.yaml @@ -51,11 +51,11 @@ models: - name: YOLOv8-Detection-Quantized performance_metrics: - torchscript_onnx_tflite: - inference_time: 2111.0 - throughput: 473.70914258645195 + inference_time: 2101.0 + throughput: 475.9638267491671 estimated_peak_memory_range: - min: 16384 - max: 1700952 + min: 12288 + max: 94625768 primary_compute_unit: NPU precision: int8 layer_info: @@ -63,7 +63,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 280 - job_id: jz5wy7lzg + job_id: jqpyyd10p job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -72,13 +72,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:23:18Z' + timestamp: '2024-08-26T23:41:44Z' - torchscript_onnx_tflite: - inference_time: 1412.0 - throughput: 708.2152974504249 + inference_time: 1410.0 + throughput: 709.2198581560284 estimated_peak_memory_range: min: 12288 - max: 54197952 + max: 53942896 primary_compute_unit: NPU precision: int8 layer_info: @@ -86,7 +86,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 280 - job_id: jmg9omzqg + job_id: j2p0xr40p job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -95,13 +95,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:23:19Z' + timestamp: '2024-08-26T23:41:45Z' - torchscript_onnx_tflite: - inference_time: 2104.0 - throughput: 475.2851711026616 + inference_time: 2092.0 + throughput: 478.0114722753346 estimated_peak_memory_range: min: 12288 - max: 2368736 + max: 1587704 primary_compute_unit: NPU precision: int8 layer_info: @@ -109,7 +109,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 280 - job_id: jnp1ojnk5 + job_id: j1p8k72qp job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -118,13 +118,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:23:20Z' + timestamp: '2024-08-26T23:41:46Z' - torchscript_onnx_tflite: - inference_time: 2349.0 - throughput: 425.7130693912303 + inference_time: 2306.0 + throughput: 433.6513443191674 estimated_peak_memory_range: - min: 12288 - max: 56914816 + min: 16384 + max: 56369184 primary_compute_unit: NPU precision: int8 layer_info: @@ -132,7 +132,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 280 - job_id: jvgd63dkp + job_id: jogkkyvvg job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -141,13 +141,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:23:21Z' + timestamp: '2024-08-26T23:41:47Z' - torchscript_onnx_tflite: - inference_time: 2125.0 - throughput: 470.5882352941176 + inference_time: 2094.0 + throughput: 477.5549188156638 estimated_peak_memory_range: min: 16384 - max: 4969968 + max: 69089136 primary_compute_unit: NPU precision: int8 layer_info: @@ -155,7 +155,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 280 - job_id: jz57o4eqg + job_id: jn5qd20eg job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -164,13 +164,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:23:22Z' + timestamp: '2024-08-26T23:41:48Z' - torchscript_onnx_tflite: - inference_time: 2117.0 - throughput: 472.3665564478035 + inference_time: 2084.0 + throughput: 479.8464491362764 estimated_peak_memory_range: - min: 16384 - max: 13037848 + min: 12288 + max: 1749312 primary_compute_unit: NPU precision: int8 layer_info: @@ -178,7 +178,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 280 - job_id: jqp4e1yqg + job_id: j1glqk42p job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -187,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:23:23Z' + timestamp: '2024-08-26T23:41:48Z' - torchscript_onnx_tflite: - inference_time: 2112.0 - throughput: 473.4848484848485 + inference_time: 2094.0 + throughput: 477.5549188156638 estimated_peak_memory_range: - min: 16384 - max: 76821200 + min: 12288 + max: 4977232 primary_compute_unit: NPU precision: int8 layer_info: @@ -201,7 +201,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 1 total_layers: 280 - job_id: j0px04ljp + job_id: jw56012n5 job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -210,13 +210,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:23:24Z' + timestamp: '2024-08-26T23:41:49Z' - torchscript_onnx_tflite: - inference_time: 6051.0 - throughput: 165.26194017517767 + inference_time: 5789.0 + throughput: 172.74140611504578 estimated_peak_memory_range: - min: 45056 - max: 37563152 + min: 86016 + max: 39157648 primary_compute_unit: NPU precision: int8 layer_info: @@ -224,7 +224,7 @@ models: layers_on_gpu: 1 layers_on_cpu: 1 total_layers: 280 - job_id: jo5m9m0yg + job_id: j1p3rmnmp job_status: Passed reference_device_info: name: RB3 Gen 2 (Proxy) @@ -233,13 +233,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs6490 - timestamp: '2024-08-10T22:23:25Z' + timestamp: '2024-08-26T23:41:50Z' - torchscript_onnx_tflite: - inference_time: 47749.0 - throughput: 20.942846970617186 + inference_time: 47463.0 + throughput: 21.069043254745804 estimated_peak_memory_range: - min: 2854912 - max: 10480824 + min: 2891776 + max: 12598048 primary_compute_unit: NPU precision: int8 layer_info: @@ -247,7 +247,7 @@ models: layers_on_gpu: 2 layers_on_cpu: 1 total_layers: 280 - job_id: jegn1n1vp + job_id: jwgo9vz1g job_status: Passed reference_device_info: name: RB5 (Proxy) @@ -256,4 +256,4 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8250 - timestamp: '2024-08-10T22:23:27Z' + timestamp: '2024-08-26T23:41:51Z' diff --git a/qai_hub_models/models/yolov8_seg/export.py b/qai_hub_models/models/yolov8_seg/export.py index 2c3bb513..1a363bba 100644 --- a/qai_hub_models/models/yolov8_seg/export.py +++ b/qai_hub_models/models/yolov8_seg/export.py @@ -217,9 +217,7 @@ def export_model( def main(): warnings.filterwarnings("ignore") - parser = export_parser( - model_cls=Model, supports_qnn=False, supports_precompiled_qnn_onnx=False - ) + parser = export_parser(model_cls=Model) args = parser.parse_args() export_model(**vars(args)) diff --git a/qai_hub_models/models/yolov8_seg/perf.yaml b/qai_hub_models/models/yolov8_seg/perf.yaml index 0b1f709c..9fa6ab0b 100644 --- a/qai_hub_models/models/yolov8_seg/perf.yaml +++ b/qai_hub_models/models/yolov8_seg/perf.yaml @@ -45,11 +45,11 @@ models: - name: YOLOv8-Segmentation performance_metrics: - torchscript_onnx_tflite: - inference_time: 6484.0 - throughput: 154.22578655151142 + inference_time: 6539.0 + throughput: 152.9285823520416 estimated_peak_memory_range: - min: 4243456 - max: 14408384 + min: 12288 + max: 37155616 primary_compute_unit: NPU precision: fp16 layer_info: @@ -57,14 +57,29 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 338 - job_id: jo5m9mm7g + job_id: jw5601yy5 + job_status: Passed + torchscript_onnx_qnn: + inference_time: 6404.0 + throughput: 156.1524047470331 + estimated_peak_memory_range: + min: 4210688 + max: 16896056 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 333 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 333 + job_id: jz5wr9emp job_status: Passed torchscript_onnx: - inference_time: 9095.0 - throughput: 109.95052226498076 + inference_time: 9138.0 + throughput: 109.4331363536879 estimated_peak_memory_range: - min: 17580032 - max: 20594304 + min: 16764928 + max: 19260888 primary_compute_unit: NPU precision: fp16 layer_info: @@ -72,7 +87,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 339 - job_id: jlpe6z78g + job_id: jz578dynp job_status: Passed reference_device_info: name: Samsung Galaxy S23 @@ -81,13 +96,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 2 - timestamp: '2024-08-10T22:22:37Z' + timestamp: '2024-08-26T23:41:03Z' - torchscript_onnx_tflite: - inference_time: 4692.0 - throughput: 213.12872975277068 + inference_time: 4652.0 + throughput: 214.96130696474634 estimated_peak_memory_range: - min: 4268032 - max: 105965664 + min: 12288 + max: 104160768 primary_compute_unit: NPU precision: fp16 layer_info: @@ -95,14 +110,29 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 338 - job_id: jegn1nnjp + job_id: j1p3rmjnp + job_status: Passed + torchscript_onnx_qnn: + inference_time: 4572.0 + throughput: 218.72265966754156 + estimated_peak_memory_range: + min: 20000768 + max: 72169392 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 333 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 333 + job_id: jmg9q4l8p job_status: Passed torchscript_onnx: - inference_time: 6613.0 - throughput: 151.21729925903523 + inference_time: 6532.0 + throughput: 153.09246785058176 estimated_peak_memory_range: - min: 19525632 - max: 131556128 + min: 20860928 + max: 131090496 primary_compute_unit: NPU precision: fp16 layer_info: @@ -110,7 +140,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 339 - job_id: jygzzml4g + job_id: jqp42wl2g job_status: Passed reference_device_info: name: Samsung Galaxy S24 @@ -119,13 +149,13 @@ models: os_name: Android manufacturer: Samsung chipset: Snapdragon® 8 Gen 3 - timestamp: '2024-08-10T22:22:38Z' + timestamp: '2024-08-26T23:41:04Z' - torchscript_onnx_tflite: - inference_time: 6477.0 - throughput: 154.3924656476764 + inference_time: 6566.0 + throughput: 152.29972586049345 estimated_peak_memory_range: - min: 4599808 - max: 26258744 + min: 4587520 + max: 14560720 primary_compute_unit: NPU precision: fp16 layer_info: @@ -133,7 +163,22 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 338 - job_id: joprx00kp + job_id: jwgo9v2kg + job_status: Passed + torchscript_onnx_qnn: + inference_time: 6161.0 + throughput: 162.31131309852296 + estimated_peak_memory_range: + min: 4988928 + max: 6337416 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 333 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 333 + job_id: jvgdmvxzg job_status: Passed reference_device_info: name: QCS8550 (Proxy) @@ -142,13 +187,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8550 - timestamp: '2024-08-10T22:22:21Z' + timestamp: '2024-08-26T23:40:59Z' - torchscript_onnx_tflite: - inference_time: 9246.0 - throughput: 108.1548777849881 + inference_time: 9320.0 + throughput: 107.29613733905579 estimated_peak_memory_range: - min: 4587520 - max: 96120704 + min: 4591616 + max: 98131392 primary_compute_unit: NPU precision: fp16 layer_info: @@ -156,7 +201,22 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 338 - job_id: jep2oww6g + job_id: j1pvnw6rg + job_status: Passed + torchscript_onnx_qnn: + inference_time: 8974.0 + throughput: 111.43302874972142 + estimated_peak_memory_range: + min: 4947968 + max: 43696944 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 333 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 333 + job_id: jvgdmvx6g job_status: Passed reference_device_info: name: QCS8450 (Proxy) @@ -165,13 +225,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Qcs8450 - timestamp: '2024-08-10T22:22:22Z' + timestamp: '2024-08-26T23:41:03Z' - torchscript_onnx_tflite: - inference_time: 6524.0 - throughput: 153.28019619865114 + inference_time: 6416.0 + throughput: 155.86034912718205 estimated_peak_memory_range: - min: 4599808 - max: 14373096 + min: 0 + max: 12009504 primary_compute_unit: NPU precision: fp16 layer_info: @@ -179,7 +239,22 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 338 - job_id: jqpy8xx0g + job_id: j7gj8lve5 + job_status: Passed + torchscript_onnx_qnn: + inference_time: 6086.0 + throughput: 164.31153466973382 + estimated_peak_memory_range: + min: 5578752 + max: 11972288 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 333 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 333 + job_id: jz5wr9e4p job_status: Passed reference_device_info: name: SA8650 (Proxy) @@ -188,13 +263,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8650p - timestamp: '2024-08-10T22:22:24Z' + timestamp: '2024-08-26T23:41:00Z' - torchscript_onnx_tflite: inference_time: 6528.0 throughput: 153.18627450980392 estimated_peak_memory_range: - min: 12288 - max: 11914376 + min: 4583424 + max: 93804584 primary_compute_unit: NPU precision: fp16 layer_info: @@ -202,7 +277,22 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 338 - job_id: j2p0ojm0p + job_id: jlpenvdvp + job_status: Passed + torchscript_onnx_qnn: + inference_time: 6152.0 + throughput: 162.5487646293888 + estimated_peak_memory_range: + min: 4972544 + max: 12795320 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 333 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 333 + job_id: jmg9q4lmp job_status: Passed reference_device_info: name: SA8775 (Proxy) @@ -211,13 +301,13 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8775p - timestamp: '2024-08-10T22:22:25Z' + timestamp: '2024-08-26T23:41:01Z' - torchscript_onnx_tflite: - inference_time: 6493.0 - throughput: 154.01201293700908 + inference_time: 6451.0 + throughput: 155.0147263990079 estimated_peak_memory_range: min: 4608000 - max: 14588496 + max: 24701856 primary_compute_unit: NPU precision: fp16 layer_info: @@ -225,7 +315,22 @@ models: layers_on_gpu: 0 layers_on_cpu: 0 total_layers: 338 - job_id: j1p8jxeq5 + job_id: jygz073x5 + job_status: Passed + torchscript_onnx_qnn: + inference_time: 6190.0 + throughput: 161.55088852988692 + estimated_peak_memory_range: + min: 4968448 + max: 10910424 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 333 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 333 + job_id: jnp1m84np job_status: Passed reference_device_info: name: SA8255 (Proxy) @@ -234,13 +339,28 @@ models: os_name: Android manufacturer: Qualcomm chipset: Sa8255p - timestamp: '2024-08-10T22:22:26Z' - - torchscript_onnx: - inference_time: 9155.0 - throughput: 109.22992900054615 + timestamp: '2024-08-26T23:41:02Z' + - torchscript_onnx_qnn: + inference_time: 6572.0 + throughput: 152.16068167985392 + estimated_peak_memory_range: + min: 4923392 + max: 4923392 + primary_compute_unit: NPU + precision: fp16 + layer_info: + layers_on_npu: 333 + layers_on_gpu: 0 + layers_on_cpu: 0 + total_layers: 333 + job_id: jnp1m847p + job_status: Passed + torchscript_onnx: + inference_time: 9106.0 + throughput: 109.81770261366133 estimated_peak_memory_range: - min: 17522688 - max: 17522688 + min: 17510400 + max: 17510400 primary_compute_unit: NPU precision: fp16 layer_info: @@ -248,7 +368,7 @@ models: layers_on_gpu: 0 layers_on_cpu: 2 total_layers: 339 - job_id: jz5wy7l4g + job_id: j0pxz1k85 job_status: Passed reference_device_info: name: Snapdragon X Elite CRD @@ -257,4 +377,4 @@ models: os_name: Windows manufacturer: Qualcomm chipset: Snapdragon® X Elite - timestamp: '2024-08-10T22:22:39Z' + timestamp: '2024-08-26T23:41:05Z' diff --git a/qai_hub_models/test/test_async_compile_jobs.py b/qai_hub_models/test/test_async_compile_jobs.py index 18b18235..8ba7c243 100644 --- a/qai_hub_models/test/test_async_compile_jobs.py +++ b/qai_hub_models/test/test_async_compile_jobs.py @@ -24,7 +24,8 @@ def test_compile_jobs_success(): timeout_jobs = {} for name, job_id in job_ids.items(): job = hub.get_job(job_id) - if job.get_status().running: + status = job.get_status() + if status.running or status.pending: # Wait a maximum of 15 minutes for a compile job timemax = datetime.timedelta(minutes=15) timediff = datetime.datetime.now() - job.date diff --git a/scripts/build_and_test.py b/scripts/build_and_test.py index d42f8b7d..9fc4beb5 100755 --- a/scripts/build_and_test.py +++ b/scripts/build_and_test.py @@ -15,6 +15,7 @@ REPRESENTATIVE_EXPORT_MODELS, get_all_models, get_changed_models, + get_code_gen_changed_models, get_models_to_run_general_tests, get_models_to_test_export, get_models_with_changed_definitions, @@ -264,48 +265,36 @@ def test_scripts(self, plan: Plan, step_id: str = "test_scripts") -> str: def test_changed_models( self, plan: Plan, step_id: str = "test_changed_models" ) -> str: - changed_model_defs = set( - get_models_with_changed_definitions() - ) # model.py changed - export_changed_models = set( - get_models_with_export_file_changes() - ) # export.py or test_generated.py changed - - # Get the set of models for which export changed and model defs changed - model_and_export_changed = changed_model_defs & export_changed_models - if len(model_and_export_changed) > 0: - # Don't bother testing all models for export. - # Just test the export for the models whose definitions changed. - export_models = model_and_export_changed - elif len(export_changed_models) > 0: - # This is true when `export.py` or `test_generated.py` are mass-changed, - # but no model definitions actually changed. That means this was a mass-change - # to the export scripts. - # - # Test a representative set of models. - # One regular model, one aimet, one components, and one non-image input. - # These are among the smallest instances of each of these. - # If none of these models were changed, test one model. - export_models = export_changed_models & set(REPRESENTATIVE_EXPORT_MODELS) - if len(export_models) == 0: - export_models = set([next(iter(export_changed_models))]) - else: - export_models = set() + # model.py changed + model_changed_models = get_models_with_changed_definitions() + + # export.py or test_generated.py changed + export_changed_models = get_models_with_export_file_changes() + + # code-gen.yaml changed + code_gen_changed_models = get_code_gen_changed_models() + + # If model or code-gen changed, then test export. + models_to_test_export = model_changed_models | code_gen_changed_models + + # For all other models where export.py or test_generated.py changed, + # only test if they're part of REPRESENTATIVE_EXPORT_MODELS + models_to_test_export.update( + export_changed_models & set(REPRESENTATIVE_EXPORT_MODELS) + ) + + # Set of models where model.py, demo.py, or test.py changed. + models_to_run_tests = get_models_to_run_general_tests() - # Set of models to run general tests - models_to_run_tests = set( - get_models_to_run_general_tests() - ) # demo.py or model.py changed - models_to_run_tests = ( - models_to_run_tests | export_models - ) # export tests can only run alongside general model tests + # export tests can only run alongside general model tests + models_to_run_tests = models_to_run_tests | models_to_test_export return plan.add_step( step_id, PyTestModelsTask( self.python_executable, models_to_run_tests, - export_models, + models_to_test_export, self.venv_path, venv_for_each_model=False, use_shared_cache=True, @@ -320,7 +309,7 @@ def test_changed_models( def test_changed_models_long( self, plan: Plan, step_id: str = "test_changed_models_long" ) -> str: - default_test_models = ["mobilenet_v2", "googlenet"] + default_test_models = REPRESENTATIVE_EXPORT_MODELS return plan.add_step( step_id, PyTestModelsTask( diff --git a/scripts/examples/quantize_mediapipe_face.py b/scripts/examples/quantize_mediapipe_face.py new file mode 100644 index 00000000..f04c2c8e --- /dev/null +++ b/scripts/examples/quantize_mediapipe_face.py @@ -0,0 +1,83 @@ +# --------------------------------------------------------------------- +# Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# --------------------------------------------------------------------- +import argparse +import os +from pathlib import Path + +import numpy as np +import torch +from tqdm import tqdm + +from qai_hub_models.models.mediapipe_face.app import MediaPipeFaceApp +from qai_hub_models.models.mediapipe_face_quantized.model import ( + MediaPipeFace, + MediaPipeFaceQuantizable, +) +from qai_hub_models.utils.asset_loaders import load_image + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument( + "--num-iter", type=int, default=8, help="Number of images to use." + ) + parser.add_argument( + "--output-dir", + type=str, + default=None, + help="Directory where encodings should be stored. Defaults to ./build.", + ) + parser.add_argument( + "--seed", + type=int, + default=42, + help="Manual seed to ensure reproducibility for quantization.", + ) + parser.add_argument( + "--dataset-dir", + type=str, + required=True, + help="Path to folder containing face images to calibrate the model.", + ) + + args = parser.parse_args() + image_files = os.listdir(args.dataset_dir) + rng = np.random.default_rng(args.seed) + rng.shuffle(image_files) + + image_files = [Path(args.dataset_dir) / fp for fp in image_files[: args.num_iter]] + + model = MediaPipeFaceQuantizable.from_pretrained( + face_detector_encodings=None, landmark_detector_encodings=None + ) + + def _calibrate_model(model: torch.nn.Module, args): + app = MediaPipeFaceApp(MediaPipeFace.from_pretrained()) + model_is_landmark_detector, image_paths = args + if model_is_landmark_detector: + app.landmark_detector = model + else: + app.detector = model + for image_path in tqdm(image_paths): + image = load_image(image_path).convert("RGB") + app.predict_landmarks_from_image(image) + + model.face_detector.quant_sim.compute_encodings( + _calibrate_model, [False, image_files] + ) + model.face_landmark_detector.quant_sim.compute_encodings( + _calibrate_model, [True, image_files] + ) + + output_path = args.output_dir or str(Path() / "build") + + model.face_detector.quant_sim.save_encodings_to_json( + output_path, "face_detector_quantized_encodings" + ) + print(f"Wrote {output_path}/face_detector_quantized_encodings.json\n") + + model.face_landmark_detector.quant_sim.save_encodings_to_json( + output_path, "landmark_detector_quantized_encodings" + ) + print(f"Wrote {output_path}/landmark_detector_quantized_encodings.json\n") diff --git a/scripts/tasks/changes.py b/scripts/tasks/changes.py index 647e758f..dc64da91 100644 --- a/scripts/tasks/changes.py +++ b/scripts/tasks/changes.py @@ -3,7 +3,8 @@ # SPDX-License-Identifier: BSD-3-Clause # --------------------------------------------------------------------- import os -from typing import Iterable +from pathlib import Path +from typing import Iterable, Optional, Set from .constants import ( PY_PACKAGE_MODELS_ROOT, @@ -58,7 +59,7 @@ def get_python_import_expression(filepath: str) -> str: return rel_path.replace("/", ".") -def _get_file_edges(filename): +def _get_file_edges(filename) -> Set[str]: """ Resolve which files directly import from `filename`. """ @@ -93,7 +94,7 @@ def resolve_affected_models( include_export: bool = True, include_tests: bool = True, include_generated_tests: bool = True, -) -> Iterable[str]: +) -> Set[str]: """ Given a list of changed python files, performs a Depth-First Search (DFS) over the qai_hub_models directory to figure out which directories were affected. @@ -125,8 +126,8 @@ def resolve_affected_models( changed_models = set() for f in seen: if f.startswith(PY_PACKAGE_RELATIVE_MODELS_ROOT): - basename = os.path.basename(f) - if basename not in [ + file_path = Path(f) + if file_path.name not in [ "model.py", "export.py", "test.py", @@ -134,28 +135,39 @@ def resolve_affected_models( "demo.py", ]: continue - if not include_model and basename == "model.py": + if not include_model and file_path.name == "model.py": continue - if not include_export and basename == "export.py": + if not include_export and file_path.name == "export.py": continue - if not include_tests and basename == "test.py": + if not include_tests and file_path.name == "test.py": continue - if not include_generated_tests and basename == "test_generated.py": + if not include_generated_tests and file_path.name == "test_generated.py": continue - if not include_demo and basename == "demo.py": + if not include_demo and file_path.name == "demo.py": continue - model_name = f[len(PY_PACKAGE_RELATIVE_MODELS_ROOT) :].split("/")[1] - if os.path.exists( - os.path.join(PY_PACKAGE_MODELS_ROOT, model_name, "model.py") - ): + model_name = file_path.parent.name + if (file_path.parent / "model.py").exists(): changed_models.add(model_name) return changed_models -def get_changed_files_in_package() -> Iterable[str]: +def get_code_gen_changed_models() -> Set[str]: + """Get models where the `code-gen.yaml` changed.""" + changed_code_gen_files = get_changed_files_in_package("code-gen.yaml") + changed_models = [] + for f in changed_code_gen_files: + if not f.startswith(PY_PACKAGE_RELATIVE_MODELS_ROOT): + continue + changed_models.append(Path(f).parent.name) + return set(changed_models) + + +def get_changed_files_in_package(suffix: Optional[str] = None) -> Iterable[str]: """ Returns the list of changed files in zoo based on git tracking. + + If the suffix argument is passed, restrict only to files ending in that suffix. """ with new_cd(REPO_ROOT): os.makedirs("build/model-zoo/", exist_ok=True) @@ -168,14 +180,14 @@ def get_changed_files_in_package() -> Iterable[str]: file for file in f.read().split("\n") if file.startswith(PY_PACKAGE_RELATIVE_SRC_ROOT) - and file.endswith(".py") + and (suffix is None or file.endswith(suffix)) ] # Weed out duplicates return list(set(changed_files)) return [] -def get_models_to_test_export() -> Iterable[str]: +def get_models_to_test_export() -> Set[str]: """ The models for which to test export (i.e. compilation to .tflite). Current heuristic is to only do this for models where model.py or @@ -190,7 +202,7 @@ def get_models_to_test_export() -> Iterable[str]: ) -def get_models_with_export_file_changes() -> Iterable[str]: +def get_models_with_export_file_changes() -> Set[str]: """ The models for which to test export (i.e. compilation to .tflite). Current heuristic is to only do this for models where model.py or @@ -205,7 +217,7 @@ def get_models_with_export_file_changes() -> Iterable[str]: ) -def get_models_with_changed_definitions() -> Iterable[str]: +def get_models_with_changed_definitions() -> Set[str]: """ The models for which to run non-generated (demo / model) tests. """ @@ -218,7 +230,7 @@ def get_models_with_changed_definitions() -> Iterable[str]: ) -def get_models_to_run_general_tests() -> Iterable[str]: +def get_models_to_run_general_tests() -> Set[str]: """ The models for which to run non-generated (demo / model) tests. """ @@ -237,7 +249,7 @@ def get_changed_models( include_export: bool = True, include_tests: bool = True, include_generated_tests: bool = True, -) -> Iterable[str]: +) -> Set[str]: """ Resolve which models within zoo have changed to figure which ones need to be tested. @@ -248,7 +260,7 @@ def get_changed_models( Returns a list of model IDs (folder names) that have changed. """ return resolve_affected_models( - get_changed_files_in_package(), + get_changed_files_in_package(".py"), include_model, include_demo, include_export, @@ -257,7 +269,7 @@ def get_changed_models( ) -def get_all_models() -> Iterable[str]: +def get_all_models() -> Set[str]: """ Resolve model IDs (folder names) of all models in QAIHM. """