lm head in the trained model is not in AWQ format
#6
by
pooya-mohammadi
- opened
Based on the following line:
https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct-AWQ/blob/main/model.safetensors.index.json#L1126
the lm-head is saved in linear format and because of that in transformers the lm_head cannot load the proper weights.
I believe this needs to be fixed on the saved models since the transformers is working properly.
Best