set support static cache for qwen2

[skaulintel]

In upstream, supporting static cache for this model is temporarily set to False (https://github.com/huggingface/transformers/blame/752ef3fd4e70869626ec70657a770a85c0ad9219/src/transformers/models/qwen2_vl/modeling_qwen2_vl.py#L941). We will set it to true here to allow the following test to run

python3 /root/optimum-habana/examples/image-to-text/run_pipeline.py --model_name_or_path Qwen/Qwen2-VL-2B-Instruct --batch_size 1 --max_new_tokens 20 --ignore_eos --use_hpu_graphs --bf16 --sdp_on_bf16 --output_dir /tmp/tmp_26vvp09

If not, we get the following value error

File "/usr/local/lib/python3.10/dist-packages/optimum/habana/transformers/generation/utils.py", line 1033, in _prepare_cache_for_generation
    raise ValueError(
ValueError: This model does not support `cache_implementation='static'`. Please check the following issue: https://github.com/huggingface/transformers/issues/28981

[malkomes]

I think you mean to say that you temporarily set it to True ;-)

[12010486]

Good for me, but could you add also a comment in the file on why we are overwriting the variable? Also a link to https://github.com/huggingface/transformers/blame/66f29aaaf55c8fe0c3dbcd24beede2ca4effac56/src/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py#L390C5-L390C27 would do the job.

Why: I'm afraid we need to keep a look on it, for when we are going to shift the default to eager/torch.compile

[12010486]

Good for me, but could you add also a comment in the file on why we are overwriting the variable? Also a link to https://github.com/huggingface/transformers/blame/66f29aaaf55c8fe0c3dbcd24beede2ca4effac56/src/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py#L390C5-L390C27 would do the job.

Why: I'm afraid we need to keep a look on it, for when we are going to shift the default to eager/torch.compile

@regisss, besides the comment, LGTM