Added support for new model databricks/dbrx-base #82
+493
−21
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Ann <[email protected]>
quic-akuruvil
requested review from
quic-mamta,
irajagop,
anujgupt-github,
vbaddi and
ochougul
as code owners
Merged
ochougul
requested changes
Aug 13, 2024
| @@ -114,6 +123,12 @@ | |||
| GPT2Block: QEffGPT2Block, | |||
| GPT2Attention: QEffGPT2Attention, | |||
| GPT2LMHeadModel: QEffGPT2LMHeadModel, | |||
| # Dbrx model layers | |||
There was a problem hiding this comment.
Please add in QEfficient/transformers/pytorch_transforms.py::KVCacheTransform too.
We will be deprecating this after 1.18 release.
|
|
||
| from QEfficient.transformers.modeling_attn_mask_utils import _create_causal_mask | ||
|
|
||
| DBRX_ATTENTION_CLASSES = { |
| @@ -219,7 +219,16 @@ def get_padding_shape_from_config(config, batch_size, seq_len): | |||
| ): # Check for num_key_value_heads (Llama/Mistral) | |||
| n_heads = config.num_key_value_heads | |||
| d_head = config.hidden_size // config.num_attention_heads | |||
| elif hasattr(config, "n_heads"): # Check for n_heads and d_model in the config (MPT Model) | |||
| elif ( | |||
There was a problem hiding this comment.
isn't this condition same as line 231-233?
We can move those line here, and remove 223-226?
There was a problem hiding this comment.
Actually no, the MPT model needed more specific parameters. Because simply testing for n_heads causes dbrx also, to satisfy the condition. MPT and dbrx has similar config.json. Hence current check fails, and wrong config will be set for the model.
irajagop
requested changes
Aug 28, 2024
Comment on lines
+92
to
+115
| # Save model to single weight file | ||
| params = sum(p.numel() for p in pt_model.parameters()) | ||
| model_size = math.ceil((params * 4) / Constants.GB) | ||
| if model_size < 380: | ||
| info("ONNX model uses external data. Saving external data as single weight file.") | ||
| loaded_model = onnx.load(f"{gen_models_path}_tmp/{model_base_name}.onnx") | ||
| os.makedirs(f"{gen_models_path}", exist_ok=True) | ||
| shutil.rmtree(f"{gen_models_path}_tmp") | ||
| info("Clearing files .. ") | ||
| onnx.save_model( | ||
| loaded_model, | ||
| os.path.join(gen_models_path, f"{model_base_name}.onnx"), | ||
| save_as_external_data=True, | ||
| all_tensors_to_one_file=True, | ||
| location=f"{model_base_name}.onnxweights.data", | ||
| size_threshold=1024, | ||
| convert_attribute=False, | ||
| ) | ||
| onnx.checker.check_model(os.path.join(gen_models_path, f"{model_base_name}.onnx")) | ||
| else: | ||
| info("Skip saving external data as a single file.") | ||
| if os.path.exists(f"{gen_models_path}"): | ||
| shutil.rmtree(f"{gen_models_path}") | ||
| shutil.move(f"{gen_models_path}_tmp", f"{gen_models_path}") |
There was a problem hiding this comment.
This can be done with SplitTensorsTransform, now merged into main. Is this change tested with other models also?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
config.jsonfile attributes