LAVIS icon indicating copy to clipboard operation
LAVIS copied to clipboard

A reminder and question about the vicuna checkpoint

Open Richar-Du opened this issue 10 months ago • 2 comments

As a reminder, I find that the config of eachadea/vicuna-7b-1.1 and lmsys-vicuna-7b-v1.1 are different, i.e. they have different bos_token_id, eos_token_id, and pad_token_id, and only eachadea/vicuna-7b-1.1 can work well with instructBLIP.

By the way, I have a question about that: why these special token_ids are different but the model weight and other files are the same between eachadea/vicuna-7b-1.1 and lmsys-vicuna-7b-v1.1? In my opinion, if the token_id is changed, the model needs to be retrained. I would be appreciated if anybody could explain this :)

Richar-Du avatar Aug 14 '23 06:08 Richar-Du

Are there any update to this issue?

linzhiqiu avatar Sep 13 '23 06:09 linzhiqiu

same question! How to deal with it??

onevae avatar Mar 19 '24 05:03 onevae