Add support transformers v4.54 v4.55 #1406
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
| @@ -4685,14 +4685,18 @@ def gpt_bigcode_attn(self, query, key, value, attention_mask=None, head_mask=Non | |||
| class GptBigCodeModelPatcher(OVDecoderModelPatcher): | |||
| def __enter__(self): | |||
| super().__enter__() | |||
| if getattr(self._model.config, "_attn_implementation", "eager") == "sdpa": | |||
There was a problem hiding this comment.
| @@ -3716,7 +3742,42 @@ def patch_model_for_export( | |||
| ], | |||
| library_name="transformers", | |||
| ) | |||
| class GPTBigCodeOpenVINOConfig(GPTBigCodeOnnxConfig): | |||
| class GPTBigCodeOpenVINOConfig(TextDecoderWithPositionIdsOnnxConfig): | |||
There was a problem hiding this comment.
following huggingface/transformers#38635, the pkv generation was also updated in GPTBigCodeDummyPastKeyValuesGenerator huggingface/optimum#2336
| return attn_output, None, past_key_value | ||
| outputs = (attn_output, None) | ||
|
|
||
| if is_transformers_version("<", "4.54"): |
There was a problem hiding this comment.
following huggingface/transformers#38635
tests/openvino/test_modeling.py
Outdated
| "bloom": 5 if is_transformers_version(">=", "4.44") else 0, | ||
| "chatglm": 2, | ||
| "codegen": 5, | ||
| "codegen2": 2, | ||
| "gpt2": 5, | ||
| "gptj": 5, | ||
| "gpt_neo": 4, | ||
| "gpt_neox": 5, | ||
| "gpt2": 5 if is_transformers_version(">=", "4.42") else 0, | ||
| "gptj": 5 if is_transformers_version(">=", "4.42") else 0, | ||
| "gpt_neo": 4 if is_transformers_version(">=", "4.42") else 0, | ||
| "gpt_neox": 5 if is_transformers_version(">=", "4.42") else 0, |
There was a problem hiding this comment.
coming from optimum bettertransformers depreciation huggingface/optimum#2305
There was a problem hiding this comment.
4.42 or 4.45 might be a good new min version
There was a problem hiding this comment.
yes agreed! upgraded to 4.45
| @@ -145,7 +146,7 @@ | |||
| "qwen2": "fxmarty/tiny-dummy-qwen2", | |||
| "qwen2_moe": "katuni4ka/tiny-random-qwen1.5-moe", | |||
| "qwen2_vl": "katuni4ka/tiny-random-qwen2vl", | |||
| "qwen2_5_vl": "katuni4ka/tiny-random-qwen2.5-vl", | |||
| "qwen2_5_vl": "optimum-internal-testing/tiny-random-qwen2.5-vl", | |||
There was a problem hiding this comment.
https://huggingface.co/katuni4ka/tiny-random-qwen2.5-vl is not compatible with transformers >= v4.55 coming from
https://github.com/huggingface/transformers/blob/v4.55.0/src/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py#L1219
optimum-internal-testing/tiny-random-qwen2.5-vl has the same parameter with out_hidden_size set to 16 to keep compatibility with transformers >= v4.55
| "transformers>=4.36,<4.54", | ||
| # "optimum==1.27.*", | ||
| "optimum-onnx @ git+https://github.com/huggingface/optimum-onnx.git@transformers454", | ||
| "transformers>=4.45,<4.56", |
There was a problem hiding this comment.
minimum transformers version now set to v4.45 which was released 1 year ago cc @IlyasMoutawwakil @helena-intel @nikita-savelyevv
|
|
||
|
|
||
| # TODO : complete | ||
| TEST_NAME_TO_MODEL_TYPE = { |
There was a problem hiding this comment.
will complete and add a tests before merging
Fix #1404