OP

OpenRLHF

High-performance RLHF framework for training LLMs

Self-Hosted AI ★ 9.9k stars Hard setup Apache-2.0

OpenRLHF is a scalable framework for reinforcement learning from human feedback, supporting PPO, DPO and other algorithms on large language models. It can be self-hosted on GPU clusters for end-to-end RLHF training.

Key features

  • PPO and DPO support
  • Distributed training
  • Ray-based scaling
  • vLLM-accelerated generation

Strengths

  • Released under the Apache-2.0 license
  • Mature project with 9.9k GitHub stars
  • Written in Python

OpenRLHF replaces

Compare OpenRLHF

13 head-to-head comparisons.

Similar self-hosted ai apps