AkariAsai / self-rag

This includes the original implementation of SELF-RAG: Learning to Retrieve, Generate and Critique through self-reflection by Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi.
https://selfrag.github.io/
MIT License
1.84k stars 171 forks source link

Questions about Critic model #60

Open leejaehoon1830 opened 8 months ago

leejaehoon1830 commented 8 months ago

I am curious whether you only used llama2 13b to create data when generating your generation model's training data, or if you also used llama2 7b to generate generation model's training data.

notoookay commented 8 months ago

Hi, If I'm correct, the authors used GPT-4 model to get the training data, check paper for more specific details.

leejaehoon1830 commented 8 months ago

I know the dataset used for training the critic model in GPT-4, but I want to know about the dataset used for the generation model.