slime

Scalable post-training framework for LLMs focused on reinforcement learning, connecting Megatron training with SGLang rollouts for flexible data generation and large-scale RL workflows.