slime-rl-training

SKILL.md

slime is an LLM post-training framework from Tsinghua's THUDM team, powering GLM-4.5, GLM-4.6, and GLM-4.7. It connects Megatron-LM for training with SGLang for high-throughput rollout generation.

Use this workflow for training reasoning models with group-relative advantages.

Use async mode for higher throughput by overlapping rollout and training.

Megatron+SGLang 프레임워크인 slime을 사용하여 RL을 통한 LLM 사후 교육에 대한 지침을 제공합니다. GLM 모델을 훈련하거나, 맞춤형 데이터 생성 워크플로를 구현하거나, RL 확장을 위해 긴밀한 Megatron-LM 통합이 필요할 때 사용하세요. 출처: orchestra-research/ai-research-skills.

원본 보기