Optimal context to provide after tokenization change
#2
by
IonizedLeaf
- opened
After seeing the ### to </s> change i was wondering if the context for vicuna is still the same of the original vicuna.
The context i adapted from the original implementation is:
A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.
### Assistant: Hello, do you need help with something?
### Human:
Is it correct to continue using ###?
This is the new setup https://github.com/lm-sys/FastChat/blob/1a204a53a388987c4faf1ecac59bbbe9c318076a/fastchat/conversation.py#L115-L124
We may need to change the C code to insert the EOS token.