New
openrlhf/openrlhf v0.11.2 — A distributed RLHF framework built on Ray and vLLM
An article summarizing the latest release v0.11.2 of the RLHF framework openrlhf/openrlhf, which supports SFT, DPO and policy-learning algorithms such as PPO, DAPO and GRPO on a Ray/vLLM-based distributed architecture, together with the scope and limitations that the source material reveals.