v0.4.0r3

Update Performance Numbers

Signed-off-by: QAIHM Team <[email protected]>

README.md

@@ -373,8 +373,8 @@ Qualcomm® AI Hub Models is licensed under BSD-3. See the [LICENSE file](../LICE
 | Model | README | Torch App | Device Export | CLI Demo
 | -- | -- | -- | -- | --
 | | | | |
-| [TrOCR](https://aihub.qualcomm.com/models/trocr) | [qai_hub_models.models.trocr](qai_hub_models/models/trocr/README.md) | ✔️ | ✔️ | ✔️
 | [OpenAI-Clip](https://aihub.qualcomm.com/models/openai_clip) | [qai_hub_models.models.openai_clip](qai_hub_models/models/openai_clip/README.md) | ✔️ | ✔️ | ✔️
+| [TrOCR](https://aihub.qualcomm.com/models/trocr) | [qai_hub_models.models.trocr](qai_hub_models/models/trocr/README.md) | ✔️ | ✔️ | ✔️
 
 ### Generative Ai
 

qai_hub_models/models/_shared/ffnet_quantized/model.py

@@ -19,11 +19,8 @@
 from aimet_torch.batch_norm_fold import fold_all_batch_norms
 from aimet_torch.model_preparer import prepare_model
 from aimet_torch.quantsim import QuantizationSimModel, load_encodings_to_sim
-from qai_hub.client import DatasetEntries
 
 from qai_hub_models.models._shared.ffnet.model import FFNet
-from qai_hub_models.utils.base_model import SourceModelFormat, TargetRuntime
-from qai_hub_models.utils.input_spec import InputSpec
 
 MODEL_ID = __name__.split(".")[-2]
 FFNET_AIMET_CONFIG = os.path.abspath(
@@ -47,14 +44,6 @@ def __init__(
         FFNet.__init__(self, ffnet_model.model)
         AIMETQuantizableMixin.__init__(self, ffnet_model)
 
-    def get_hub_compile_options(
-        self, target_runtime: TargetRuntime, other_compile_options: str = ""
-    ) -> str:
-        compile_options = super().get_hub_compile_options(
-            target_runtime, other_compile_options
-        )
-        return compile_options + " --quantize_full_type int8 --quantize_io"
-
     @classmethod
     def default_aimet_encodings(cls) -> str:
         raise NotImplementedError()
@@ -88,14 +77,3 @@ def from_pretrained(
 
         sim.model.eval()
         return cls(sim)
-
-    def preferred_hub_source_model_format(
-        self, target_runtime: TargetRuntime
-    ) -> SourceModelFormat:
-        return SourceModelFormat.ONNX
-
-    def get_calibration_data(
-        self, target_runtime: TargetRuntime, input_spec: InputSpec | None = None
-    ) -> DatasetEntries | None:
-        # Do not provide calibration data
-        return None

qai_hub_models/models/_shared/whisper/model.py

@@ -82,14 +82,6 @@ def get_input_spec() -> InputSpec:
     def from_pretrained(cls):
         return Whisper.from_pretrained().encoder
 
-    def get_hub_compile_options(
-        self, target_runtime: TargetRuntime, other_compile_options: str = ""
-    ) -> str:
-        compile_options = super().get_hub_compile_options(
-            target_runtime, other_compile_options
-        )
-        return compile_options + " --compute_unit gpu"
-
     def get_hub_profile_options(
         self, target_runtime: TargetRuntime, other_profile_options: str = ""
     ) -> str:

qai_hub_models/models/aotgan/export.py

@@ -197,7 +197,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/common.py

@@ -11,6 +11,7 @@
 class TargetRuntime(Enum):
     TFLITE = 0
     QNN = 1
+    ORT = 2
 
     def __str__(self):
         return self.name.lower()

qai_hub_models/models/convnext_tiny/export.py

@@ -189,7 +189,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ddrnet23_slim/export.py

@@ -197,7 +197,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/deeplabv3_resnet50/export.py

@@ -193,7 +193,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/densenet121/export.py

@@ -189,7 +189,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/detr_resnet101/export.py

@@ -189,7 +189,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/detr_resnet101_dc5/export.py

@@ -189,7 +189,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/detr_resnet50/export.py

@@ -189,7 +189,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/detr_resnet50_dc5/export.py

@@ -189,7 +189,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/efficientnet_b0/export.py

@@ -189,7 +189,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/esrgan/export.py

@@ -193,7 +193,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/facebook_denoiser/export.py

@@ -184,7 +184,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/fastsam_s/export.py

@@ -199,7 +199,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/fastsam_x/export.py

@@ -199,7 +199,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/fcn_resnet50/export.py

@@ -197,7 +197,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ffnet_122ns_lowres/export.py

@@ -193,7 +193,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ffnet_40s/export.py

@@ -193,7 +193,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ffnet_40s_quantized/export.py

@@ -202,7 +202,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ffnet_54s/export.py

@@ -193,7 +193,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ffnet_54s_quantized/export.py

@@ -202,7 +202,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ffnet_78s/export.py

@@ -193,7 +193,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ffnet_78s_lowres/export.py

@@ -193,7 +193,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/ffnet_78s_quantized/export.py

@@ -202,7 +202,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/googlenet/export.py

@@ -189,7 +189,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/googlenet_quantized/export.py

@@ -200,7 +200,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/googlenet_quantized/model.py

@@ -20,7 +20,6 @@
 from qai_hub_models.models.googlenet.model import GoogLeNet
 from qai_hub_models.utils.aimet.config_loader import get_default_aimet_config
 from qai_hub_models.utils.asset_loaders import CachedWebModelAsset
-from qai_hub_models.utils.base_model import SourceModelFormat, TargetRuntime
 from qai_hub_models.utils.quantization_aimet import tie_aimet_observer_groups
 
 MODEL_ID = __name__.split(".")[-2]
@@ -44,11 +43,6 @@ def __init__(
             sim_model,
         )
 
-    def preferred_hub_source_model_format(
-        self, target_runtime: TargetRuntime
-    ) -> SourceModelFormat:
-        return SourceModelFormat.ONNX
-
     @classmethod
     def from_pretrained(
         cls,

qai_hub_models/models/hrnet_pose/export.py

@@ -116,7 +116,7 @@ def export_model(
     model_compile_options = model.get_hub_compile_options(
         target_runtime,
         compile_options
-        + " --force_channel_last_input image_tensor"
+        + " --force_channel_last_input image"
         + " --force_channel_last_output output_0",
     )
     print(f"Optimizing model {model_name} to run on-device")
@@ -156,7 +156,7 @@ def export_model(
         sample_inputs = model.sample_inputs(input_spec)
         # Convert inputs from channel first to channel last
         hub_inputs = transpose_channel_first_to_last(
-            "image_tensor", sample_inputs, target_runtime
+            "image", sample_inputs, target_runtime
         )
         submitted_inference_job = hub.submit_inference_job(
             model=compile_job.get_target_model(),
@@ -197,7 +197,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model)
+    parser = export_parser(model_cls=Model, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/hrnet_pose_quantized/export.py

@@ -121,7 +121,7 @@ def export_model(
     model_compile_options = model.get_hub_compile_options(
         target_runtime,
         compile_options
-        + " --force_channel_last_input image_tensor"
+        + " --force_channel_last_input image"
         + " --force_channel_last_output output_0",
     )
     print(f"Optimizing model {model_name} to run on-device")
@@ -165,7 +165,7 @@ def export_model(
             hub_inputs = get_qnn_inputs(compile_job, sample_inputs)
         # Convert inputs from channel first to channel last
         hub_inputs = transpose_channel_first_to_last(
-            "image_tensor", sample_inputs, target_runtime
+            "image", sample_inputs, target_runtime
         )
         submitted_inference_job = hub.submit_inference_job(
             model=compile_job.get_target_model(),
@@ -206,7 +206,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))
 

qai_hub_models/models/huggingface_wavlm_base_plus/export.py

@@ -180,7 +180,7 @@ def export_model(
 
 def main():
     warnings.filterwarnings("ignore")
-    parser = export_parser(model_cls=Model, supports_qnn=False)
+    parser = export_parser(model_cls=Model, supports_qnn=False, supports_ort=False)
     args = parser.parse_args()
     export_model(**vars(args))