Shypd.aiThe largest open index of AI toolsMOSS-RLHF is an open-source research tool that explores the secrets of Reinforcement Learning from Human Feedback (RLHF) in large language models. It implements the PPO algorithm and provides code for training reward models and policy models.
Trending
About
MOSS-RLHF is an open-source project from OpenLMLab that delves into the intricacies of Reinforcement Learning from Human Feedback (RLHF) within large language models, specifically focusing on the Proximal Policy Optimization (PPO) algorithm. The project received the best paper award at the NIPS 2023 Workshop on Instruction Tuning and Instruction Following. It provides researchers with competitive Chinese and English reward models, which boast good cross-model generalization abilities, reducing the need for extensive human preference data relabeling. The project also offers in-depth analysis of the PPO algorithm, proposing the PPO-max algorithm for stable model training, and releases complete PPO-max codes to help align LLMs with human preferences. It includes resources for training reward models and policy models, along with annotated datasets.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
