Xiaohongshu Open Sources Large Model Reinforcement Learning Training Engine Relax

On April 15, the Xiaohongshu AI Platform Team officially open-sourced Relax—a large model reinforcement learning training engine designed for full-modal and Agentic scenarios. Introduced as supporting unified processing and flexible integration of inputs such as images, audio, and video, it combines modal-aware parallelism with end-to-end asynchronous pipelines to improve the efficiency and scalability of multi-modal training.