questions are from https://huggingface.co/datasets/truthful_qa
retrieved passages are from https://huggingface.co/datasets/wikipedia
generator that I'm planning to use https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B
critic that I'm planning to use https://huggingface.co/allenai/digital-socrates-7b
I also tried to fine-tune the critic with sherpherd dataset. This is my model https://huggingface.co/ShuyiGuo/allen_critic
This is Sherpherd dataset https://github.com/facebookresearch/Shepherd/tree/main/data