[BUG] <title>Error while running `Chat with video`

[ShuyUSTC]

是否已有关于该错误的issue或讨论？ | Is there an existing issue / discussion for this?

• [X] 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions

该问题是否在FAQ中有解答？ | Is there an existing answer for this in FAQ?

• [X] 我已经搜索过FAQ | I have searched FAQ

当前行为 | Current Behavior

When running the code in Chat with video https://github.com/OpenBMB/MiniCPM-V/blob/ab1141ee450ee1a3b45b3005eb1f5d4e8811b7eb/README.md?plain=1#L1416-L1469 using the video v_4LF0hL-mgks.mp4 and question: "is the person first appears in the video outdoor" in VCGBench, I have got the following traceback:

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Traceback (most recent call last):
  File "/**/chat_with_video.py", line 41, in <module>
    answer = model.chat(
  File "/~/.cache/huggingface/modules/transformers_modules/openbmb/MiniCPM-V-2_6/4e4be000cd81feda8b96d14b53f1791b4010b038/modeling_minicpmv.py", line 348, in chat
    inputs = processor(
  File "/~/.cache/huggingface/modules/transformers_modules/openbmb/MiniCPM-V-2_6/4e4be000cd81feda8b96d14b53f1791b4010b038/processing_minicpmv.py", line 67, in __call__
    return self._convert_images_texts_to_inputs(image_inputs, text, max_slice_nums=max_slice_nums, use_image_id=use_image_id, max_length=max_length, **kwargs)
  File "/~/.cache/huggingface/modules/transformers_modules/openbmb/MiniCPM-V-2_6/4e4be000cd81feda8b96d14b53f1791b4010b038/processing_minicpmv.py", line 165, in _convert_images_texts_to_inputs
    input_ids, image_bounds = self._convert(final_text, max_length)
  File "/~/.cache/huggingface/modules/transformers_modules/openbmb/MiniCPM-V-2_6/4e4be000cd81feda8b96d14b53f1791b4010b038/processing_minicpmv.py", line 121, in _convert
    image_bounds = torch.hstack(
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 124 but got size 123 for tensor number 1 in the list.

Could you please help me to solve the problem? Thaks!

期望行为 | Expected Behavior

Successfully return the anwser.

复现方法 | Steps To Reproduce

1. In the environment of MiniCPM
2. Run the code in https://github.com/OpenBMB/MiniCPM-V/blob/ab1141ee450ee1a3b45b3005eb1f5d4e8811b7eb/README.md?plain=1#L1416-L1469
3. change the video file to [v_4LF0hL-mgks.mp4](https://drive.google.com/file/d/1osAM0waEP23fvHtrIcgsdqfFkUgO hlq9/view?usp=sharing) (can be downloaded clicking the link) and question: "is the person first appears in the video outdoor"
4. See error:

Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Traceback (most recent call last):
  File "/mnt/afs_intern/shuyan/experiments/MiniCPM-V/video_chat.py", line 41, in <module>
    answer = model.chat(
  File "/mnt/afs/shuyan/.cache/huggingface/modules/transformers_modules/openbmb/MiniCPM-V-2_6/4e4be000cd81feda8b96d14b53f1791b4010b038/modeling_minicpmv.py", line 348, in chat
    inputs = processor(
  File "/mnt/afs/shuyan/.cache/huggingface/modules/transformers_modules/openbmb/MiniCPM-V-2_6/4e4be000cd81feda8b96d14b53f1791b4010b038/processing_minicpmv.py", line 67, in __call__
    return self._convert_images_texts_to_inputs(image_inputs, text, max_slice_nums=max_slice_nums, use_image_id=use_image_id, max_length=max_length, **kwargs)
  File "/mnt/afs/shuyan/.cache/huggingface/modules/transformers_modules/openbmb/MiniCPM-V-2_6/4e4be000cd81feda8b96d14b53f1791b4010b038/processing_minicpmv.py", line 165, in _convert_images_texts_to_inputs
    input_ids, image_bounds = self._convert(final_text, max_length)
  File "/mnt/afs/shuyan/.cache/huggingface/modules/transformers_modules/openbmb/MiniCPM-V-2_6/4e4be000cd81feda8b96d14b53f1791b4010b038/processing_minicpmv.py", line 121, in _convert
    image_bounds = torch.hstack(
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 124 but got size 123 for tensor number 1 in the list.

运行环境 | Environment

- OS:Ubuntu
- Python: 3.10.14
- Transformers: 4.40.0
- PyTorch:2.1.2
- CUDA (`python -c 'import torch; print(torch.version.cuda)'`): 12.1

备注 | Anything else?

No response

[wingvortex]

hi, I had the same error. Change params["max_slice_nums"] = 2 to 1 seems to work. It is suggested by the author that # use 1 if cuda OOM and video resolution > 448*448

It is also mentioned in a similar issue that the video resolution should be limited. https://github.com/OpenBMB/MiniCPM-V/issues/509

[ShuyUSTC]

Yes, it's true that change max_slice_nums to 1 avoid the error.

[expection1985]

marked