Awesome RLHF
A curated collection of datasets, models, Spaces, and papers on Reinforcement Learning from Human Feedback (RLHF).
Running142📊Note A multi-turn evaluation benchmark for chatbots which uses GPT-4 as a judge to evaluate the quality of responses.
garage-bAInd/Open-Platypus
Viewer • Updated • 17k • 335Note A high quality blend of human and synthetic datasets focused on dialogues and reasoning abilities. A good source for SFT and PPO.
meta-llama/Llama-2-7b-chat-hf
Text Generation • Updated • 1.49M • 3.47kNote The first series of open access models trained at scale using RLHF. Based on https://huggingface.co/papers/2307.09288
meta-llama/Llama-2-70b-chat-hf
Text Generation • Updated • 110k • 2.1kNote The first series of open access models trained at scale using RLHF. Based on https://huggingface.co/papers/2307.09288
lightonai/alfred-40b-0723
Text Generation • Updated • 24 • 44Note An RLHF tuned version of Falcon 40B
Anthropic/hh-rlhf
Viewer • Updated • 78.2k • 1.02kNote A dataset of dialogues between human annotators and a 52B parameter language model from Anthropic. Contains "helpfulness" and "harmlessness" subsets that can be used for training reward models. Basis for https://huggingface.co/papers/2112.00861 and https://huggingface.co/papers/2204.05862
Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
Paper • 2204.05862 • Published • 2A General Language Assistant as a Laboratory for Alignment
Paper • 2112.00861 • Published • 2stanfordnlp/SHP
Viewer • Updated • 2.92k • 268openai/summarize_from_feedback
Viewer • Updated • 4.97k • 167openai/webgpt_comparisons
Viewer • Updated • 2.35k • 203