Miles is an enterprise RL framework for LLM and VLM post-training. It connects SGLang rollout with Megatron training, supports FP8 mixed precision throughout the pipeline, and handles MoE architectures natively.
Miles is an enterprise RL framework for LLM and VLM post-training. It connects SGLang rollout with Megatron training, supports FP8 mixed precision throughout the pipeline, and handles MoE architectures natively.