slime-rl-training

What problem does it solve?

This Skill provides a robust framework for post-training Large Language Models (LLMs) using Reinforcement Learning (RL), specifically designed for scaling with frameworks like Megatron-LM and SGLang.

Core Features & Use Cases

• RL Training: Implements advanced RL algorithms like GRPO for fine-tuning LLMs.
• Megatron-LM Integration: Leverages Megatron-LM for efficient distributed training.
• SGLang for Rollouts: Utilizes SGLang for high-throughput generation during the RL process.
• Use Case: Fine-tune a Llama 3 model using RLHF to improve its conversational abilities and adherence to safety guidelines, leveraging distributed training for faster iteration.

Quick Start

Use the slime-rl-training skill to launch GRPO training with the Qwen3-4B model using the provided training data.