My observation over the previous model
#2
by
samjack
- opened
I think the biggest change is the change in the eos token.
The previous model was creating and responding by itself as if auto gpt was installed.
In this model, it stops after answering. Much more suitable for chat mode, but sad for those who like to use this model in notebook mode.
Yeah.
I've moved the legacy code to https://huggingface.co/eachadea/legacy-vicuna-13b