SLIME User Guide

SLIME is an LLM post-training framework for RL Scaling developed by THUDM. It supports various RL algorithms (GRPO, GSPO, PPO, Reinforce++), multiple training backends (Megatron, FSDP), and advanced features like multi-turn interactions, tool calling, and dynamic sampling.

Quick Start Workflow

For First-Time Users

1. Environment Setup

   • Use Docker: docker pull slimerl/slime:latest
   • Or build from source: See docs/en/get_started/quick_start.md
   • Hardware: Supports H100/H200, B200 series

2. Download Model and Data

   hf download Qwen/Qwen3-4B --local-dir /root/Qwen3-4B
   hf download --repo-type dataset zhuzilin/dapo-math-17k --local-dir /root/dapo-math-17k