skills.rest
uv-xiao
uv-xiaoCOMMUNITY
Explore all skills by this author →
Website𝕏@youwei_xiaoGitHub

uv-slime-rl-training

Run GRPO reinforcement learning training for LLMs with Megatron-LM and SGLang.

1|Updated Feb 6, 2026
Search Tags:#reinforcement learning#sglang#rlhf#megatron-lm#llm training#model fine-tuning
One-click install
npx skills add https://github.com/uv-xiao/pkbllm --skill uv-slime-rl-training
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: uv-slime-rl-training
Source: https://github.com/uv-xiao/pkbllm/tree/main/knowledge/ML/training/slime
Command: npx skills add https://github.com/uv-xiao/pkbllm --skill uv-slime-rl-training

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill 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 Megatron-LM for efficient training and SGLang for high-throughput inference.

Core Features & Use Cases

  • RL Training: Implements advanced RL algorithms like GRPO for fine-tuning LLMs.
  • Megatron-LM Integration: Leverages Megatron-LM for distributed training, supporting various parallelism strategies.
  • SGLang Rollout: Utilizes SGLang for fast and scalable response generation during the RL loop.
  • Custom Data Workflows: Supports flexible data buffering and custom generation logic for multi-turn or tool-using agents.
  • Use Case: You are training a new generation of GLM models and need a robust framework that can handle large-scale RL fine-tuning, integrating with existing Megatron-LM infrastructure and requiring high-throughput response generation for reward calculation.

Quick Start

Use the uv-slime-rl-training skill to launch GRPO training for a Qwen3-4B model using a specified data file.

Frequently Asked Questions about uv-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?▼

Reinforcement learning for LLM post-training with Megatron-LM is facilitated by the slime framework, which integrates distributed training and high-throughput inference to streamline GRPO fine-tuning.

Can I train multi-turn agents using SGLang for response generation?▼

SGLang for response generation supports multi-turn agent training by providing fast and scalable rollouts during the RL loop, enabling flexible data buffering and custom generation logic.

Does this RL training framework support model families beyond Llama?▼

RL training supports large language models beyond Llama, specifically targeting the GLM, Qwen, and DeepSeek families to ensure scalable and efficient fine-tuning.

How do I launch GRPO training for a Qwen3-4B model?▼

To launch GRPO training for a Qwen3-4B model, you use the provided skill scripts to initiate the process with a specified data file within the slime framework.

What is the benefit of using SGLang for inference during RL fine-tuning?▼

SGLang for inference during RL fine-tuning provides high-throughput response generation, which is essential for efficient reward calculation and scalable rollout in the RL loop.

Are there limitations when integrating custom data workflows for tool-using agents?▼

Custom data workflows for tool-using agents are fully supported, allowing flexible data buffering and custom generation logic, though scalable distributed training via Megatron-LM is required for large-scale execution.

Related Skills

Explore More Software Engineering Skills→
gagan114662gagan114662

verl-rl-training

Run large-scale GRPO and PPO training for LLMs with verl.

Community
Advanced
vasilyu1983vasilyu1983
73

ai-llm-inference

Optimize LLM inference performance and cost with vLLM and quantization.

Community
Advanced
matthew-johnsonmatthew-johnson

slime-rl-training

Automate RL post-training workflows for large language models with slime.

Community
Advanced
zangjeicyzangjeicy

slime-rl-training

Train LLMs with GRPO and PPO using Megatron-LM and SGLang.

Community
Advanced