tangqiaoyu/ToolAlpaca

How to use the data train LLMs?

lei1993 opened this issue · 1 comments

How to use the data train LLMs?

I apologize for the delayed response. To utilize the data for training LLMs, you would need to design a custom prompt. In our case, we adopted a prompt structure similar to LangChain MRKL. We then perform standard instruction finetuning on Vicuna models and only optimize the loss on thought, action, and action input parts.