RadixArk Launches Miles v0.1 Open-Source RL Framework

RadixArk announced today the launch of Miles v0.1, an open-source reinforcement learning framework for LLMs and multimodal models. The company states that RL training starts easily but proves hard to debug and scale, and that Miles helps verify run correctness, improve hardware efficiency, and sustain large-scale operations. Over the past nine months, 72 contributors completed 1,326 commits and 85 GPU end-to-end tests. NebiusAI reported using the framework to simplify RL workloads at scale. SGLang serves as the native rollout engine.