Hugging Face has introduced Miles v0.1, a full-stack, production-ready system designed for frontier-scale post-training and reinforcement learning (RL). Built on the clean design of slime, Miles emphasizes verified, clean, and customizable components across every stage of the RL training loop, aiming to make advanced RL accessible to both researchers and enterprises.
The system integrates a rollout engine based on SGLang, a trainer supporting two backends (NVIDIA Megatron-LM and PyTorch FSDP), and three weight-synchronization transports tailored for different deployment topologies. Beyond full-parameter RL, Miles supports LoRA RL, on-policy distillation, supervised fine-tuning, and true-on-policy rollout-training alignment, extending the same architecture to diffusion models.
In an end-to-end case study, Miles demonstrated fully asynchronous agentic RL on a GLM-5.2 744B-A40B model for terminal-use coding tasks, running on 64 NVIDIA GB300 GPUs with a median step time of 263 seconds over the first 30 measured steps.
Miles is open-sourced at https://github.com/radixark/miles, with the project website at https://miles.radixark.com.