THUDM

/

cogvlm2-llama3-chat-19B

like 194

Hi, CogVLM team,

thank you for your great work.

I looked at the inference code and there is a restriction on inputs_embeds as input. I think this affects the custom inference process. I hope to modify the assertion here. If it is acceptable, I can also submit a pull request.

https://huggingface.co/THUDM/cogvlm2-llama3-chat-19B/blob/2bf7de6892877eb50142395af14847519ba95998/modeling_cogvlm.py#L384