An asynchronous reinforcement-learning post-training engine for omni-modal large models, built on Ray Serve, Megatron-LM, and SGLang with PPO/GRPO and agentic RL support.
676agentic-rl,distributed-training,grpoAn asynchronous reinforcement-learning post-training engine for omni-modal large models, built on Ray Serve, Megatron-LM, and SGLang with PPO/GRPO and agentic RL support.