fix-qwen2vl-no-position_ids

[simonJJJ]

What does this PR do?

Fixes # (issue) (issue)

We encountered some issues with position_ids==None when fine-tuning the qwen2vl model, because the model_inputs do not pass the position_ids arg.

Before submitting

• [] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[ArthurZucker]

Makes sense! Do you want to add a small test for training? 🤗

[zhangfaen]

I am trying to finetune Qwen2-VL for Object detection on coco dataset ( https://github.com/zhangfaen/finetune-Qwen2-VL )

Without this PR, my finetuned model performance is bad.
With this PR (by cherry picking), my finetuned model performance is really good.

Any progress for this PR to be merged?

Thank you

[zytx121]

I am trying to finetune Qwen2-VL for Object detection on coco dataset ( https://github.com/zhangfaen/finetune-Qwen2-VL )

Without this PR, my finetuned model performance is bad. With this PR (by cherry picking), my finetuned model performance is really good.

Any progress for this PR to be merged?

Thank you

@zhangfaen Strange, after I modified this code, the fine-tuning performance still did not improve. Is there anything else that needs to be modified?

[zhangfaen]

I am trying to finetune Qwen2-VL for Object detection on coco dataset ( https://github.com/zhangfaen/finetune-Qwen2-VL )
Without this PR, my finetuned model performance is bad. With this PR (by cherry picking), my finetuned model performance is really good.
Any progress for this PR to be merged?
Thank you

@zhangfaen Strange, after I modified this code, the fine-tuning performance still did not improve. Is there anything else that needs to be modified?

for what task and against what dataset are you finetuning qwen2-vl ?

[simonJJJ]

hi @ArthurZucker, would u mind taking a look at merging this PR? The community is waiting for it.

[zhangfaen]

hi @ArthurZucker, would u mind taking a look at merging this PR? The community is waiting for it.

+1 @ArthurZucker

[ArthurZucker]

Yep sorry I approved I don't know why it was lost in tracks!

[Betty-J]

Yep sorry I approved I don't know why it was lost in tracks!

Hi there, when I used pip install transformers, I installed version 4.46.2, which still does not include the updates mentioned here https://github.com/huggingface/transformers/blob/main/src/transformers/models/qwen2_vl/modeling_qwen2_vl.py#L1723.