Open bigdante opened 8 months ago
Nice work. I am trying to use fastchat to train a mistral model. however, I wonder why the following code is hard code for only vicuna. https://github.com/lm-sys/FastChat/blob/main/fastchat/train/train.py
conv = get_conversation_template("vicuna") and assert conv.sep_style == SeparatorStyle.ADD_COLON_TWO
Did I misunderstand something?
Hi @bigdante After training, can you get the added_tokens.json, config.json and other json files? If so, could you share your script or code?
I wrote a patch to support dolphin format. Will share it via a PR soon, but feel free to use dolphin format for mistral in the meantime: here
Nice work. I am trying to use fastchat to train a mistral model. however, I wonder why the following code is hard code for only vicuna. https://github.com/lm-sys/FastChat/blob/main/fastchat/train/train.py
conv = get_conversation_template("vicuna") and assert conv.sep_style == SeparatorStyle.ADD_COLON_TWO
Did I misunderstand something?