prompt format?

#4
by 2EyeGuy - opened

I love your work. These uncensored models are much needed.

I'm getting tired of asking this about every model... but what prompt format was it trained on?

For goodness sake people, when you release an instruction-tuned model, you need to tell us what prompt format was used to feed instructions to the model during training. It's ridiculous that we have to guess or hunt through dozens of pages of source code searching for clues.

I assume this one was trained with the Vicuna 1.1 prompt format. But it would be nice if people would say so on the model card.

Cognitive Computations org

The model card for JuneLee/wizard-vicuna-13b should have all the information needed .

Mine is exactly that but uncensored.

https://huggingface.co/junelee/wizard-vicuna-13b/discussions/1

You can use the normal Vicuna 1.1 format:
USER: What is 4x8?
ASSISTANT:

A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: Hello! ASSISTANT: Hi!USER: How are you? ASSISTANT:

https://github.com/lm-sys/FastChat/blob/a47b8f9e93c8b5a85e81d1ae33e3a1106d8cdf80/fastchat/conversation.py#L662-L667

How to use Chat history with this Prompt template

Sign up or log in to comment