nm-testing

/

Qwen2.5-VL-7B-Instruct-FP8-Dynamic

like 0

Follow

NM Testing 54

The issue that this PR fixes is similar to what was discussed here: https://github.com/huggingface/transformers/issues/36193

Thanks for the PR, I didn't know about this issue but it makes sense given the current focus on fast image processors. Would it be simpler to just remove image_processor_type entirely and rely on HF to set the latest default if it changes again? I can apply the right change to our other models

Thanks Michael for the quick reply!

I like your idea. I just tested removing image_processor_type and ran the model via vLLM and it worked.

Would you like me to make the change in this PR?

Thanks, this was covered in https://huggingface.co/nm-testing/Qwen2.5-VL-7B-Instruct-FP8-Dynamic/discussions/3#67bbf332b24be20e87a38ce5