video_auto_processing.py breaks everything

[lucasjinreal]

System Info

the video preprocessor now breaks everytthing,

I have a model which need Qwen2.5 VL preporssor, it's OK before, but now it can not load, breaks.

My model just need Qwen2.5 VL preprocessor, it doens't have Qwen2.5 VL arch, it's a customized model.
But now, the video prerproecssor will reading the config arch in my model, rather than just the preprocessor.config.json

This is weired.

You have video processor config saved in preprocessor.json file which is deprecated. Video processor configs should be saved in their own video_preprocessor.json file. You can rena
me the file or load and save the processor back which renames it automatically. Loading from preprocessor.json will be removed in v5.0.

from transformers import AutoProcessor, Qwen2_5_VLProcessor

processor_path = './checkpoints/Qwen3-VL-2B-Unofficial'
a = Qwen2_5_VLProcessor.from_pretrained(
                processor_path, trust_remote_code=True
            )
print(a)

This code can be used without config.json must be a registered model, but now, it need confgi.sjon must be a reigstered mdoel, even though I just need preprocess_config.json

Please remove the video processor code, it's useless and introduce many bug!

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

from transformers import AutoProcessor, Qwen2_5_VLProcessor

processor_path = './checkpoints/Qwen3-VL-2B-Unofficial'
a = Qwen2_5_VLProcessor.from_pretrained(
processor_path, trust_remote_code=True
)
print(a)

Expected behavior

from transformers import AutoProcessor, Qwen2_5_VLProcessor

processor_path = './checkpoints/Qwen3-VL-2B-Unofficial'
a = Qwen2_5_VLProcessor.from_pretrained(
processor_path, trust_remote_code=True
)
print(a)

[zucchini-nlp]

Fixed by #38840

[lucasjinreal]

Will 4.53.1 introduce this fix? Also, does it still will need config.json to load the video processor when users might have a customized model and only need image preprocessor?

[zucchini-nlp]

Yes, it will be in the next v4.53 release and will not require to have config.json saved. For BC we'll check if there's an image_processor config first