slime-rl-training

Train LLMs with reinforcement learning using Megatron-LM and SGLang.

3|Updated Feb 21, 2026
One-click install
npx skills add https://github.com/ihatesea69/HieuNghi-AI-Skills --skill slime-rl-training-ihatesea69
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/ihatesea69/HieuNghi-AI-Skills/tree/main/airesearch_skills/06-post-training/slime
Command: npx skills add https://github.com/ihatesea69/HieuNghi-AI-Skills --skill slime-rl-training-ihatesea69

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sglang-router>=0.2.3, ray, torch>=2.0.0, transformers>=4.40.0, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill streamlines the complex process of post-training Large Language Models (LLMs) using Reinforcement Learning (RL), specifically integrating powerful frameworks like Megatron-LM and SGLang for efficient and scalable training.

Core Features & Use Cases

  • Integrated Framework: Combines Megatron-LM for distributed training with SGLang for high-throughput inference and data generation.
  • Flexible Data Handling: Supports custom data generation workflows and flexible prompt management.
  • Model Support: Compatible with a range of LLMs including GLM, Qwen, DeepSeek, and Llama.
  • Use Case: Researchers can leverage this Skill to fine-tune a Llama 3 model using Proximal Policy Optimization (PPO) with custom reward signals, significantly accelerating the RLHF (Reinforcement Learning from Human Feedback) process.

Quick Start

Use the slime-rl-training skill to launch GRPO training for a Qwen3 model using the provided data file.

Frequently Asked Questions about slime-rl-training

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I use reinforcement learning for LLM post-training with Megatron-LM and SGLang?

You can perform reinforcement learning LLM post-training by using the slime framework, which integrates Megatron-LM for distributed training and SGLang for efficient rollout generation. This combination streamlines scalable RLHF processes.

Can I fine-tune Qwen or Llama models using PPO with custom reward signals?

Yes, you can fine-tune model architectures like Qwen, Llama, GLM, and DeepSeek using Proximal Policy Optimization (PPO). The framework supports custom reward signals to accelerate your Reinforcement Learning from Human Feedback.

How does SGLang integrate with distributed training frameworks for LLM rollout generation?

SGLang integrates with Megatron-LM to provide high-throughput inference and efficient rollout generation during LLM reinforcement learning. This synergy allows flexible prompt management and custom data generation workflows.

Do I need Ray and PyTorch to run GRPO training for large language models?

Yes, running GRPO training requires dependencies including Ray for distributed computing and PyTorch (version 2.0.0 or higher). You also need the Transformers library and sglang-router to facilitate the LLM training process.

What is the best way to scale reinforcement learning fine-tuning for large language models?

The best way to scale LLM reinforcement learning fine-tuning is combining Megatron-LM for distributed training with SGLang for high-throughput inference. This integrated framework handles custom data workflows efficiently across various model architectures.