skills.rest
MesferAli
MesferAliCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Train GLM models with RL using Megatron-LM and SGLang.

Updated Jan 12, 2026
Search Tags:#llm#training#reinforcement learning#glm#sglang#megatron-lm
One-click install
npx skills add https://github.com/MesferAli/XCircle --skill slime-rl-training-mesferali
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/MesferAli/XCircle/tree/main/.claude/skills/slime
Command: npx skills add https://github.com/MesferAli/XCircle --skill slime-rl-training-mesferali

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sglang-router, ray, torch, transformers, 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), enabling efficient scaling and customization for models like GLM.

Core Features & Use Cases

  • RL Training Framework: Integrates Megatron-LM for distributed training with SGLang for high-throughput inference.
  • Customizable Workflows: Supports custom data generation and flexible prompt management for tailored training.
  • Use Case: Train a new version of a GLM model using Reinforcement Learning from Human Feedback (RLHF) with custom reward signals, leveraging distributed training for faster iteration.

Quick Start

Launch GRPO training for a Qwen3-4B model using a specified data file and model configuration.

Frequently Asked Questions about slime-rl-training

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

FAQPage Schema
How do I scale reinforcement learning training for LLMs with Megatron-LM?▼

To scale LLM reinforcement learning training, this framework integrates Megatron-LM for distributed training with SGLang to provide high-throughput inference. It enables asynchronous training and multi-turn agentic training to efficiently scale RL workflows.

Can I use custom reward models for RLHF with GLM models?▼

Yes, you can implement custom reward models for Reinforcement Learning from Human Feedback (RLHF) to train GLM models. The framework supports custom data generation and flexible prompt management to apply tailored reward signals during training.

What is the best way to run GRPO training for a Qwen3-4B model?▼

The best way to run GRPO training for a Qwen3-4B model is using the provided quick start scripts. You can launch training by specifying your data file and model configuration to initiate the reinforcement learning workflow.

Does this RL training framework support asynchronous training and multi-turn agentic workflows?▼

Yes, the RL training framework supports both asynchronous training and multi-turn agentic training. These advanced features are facilitated through tight Megatron-LM integration and SGLang inference to optimize complex training cycles.

Do I need Ray and PyTorch to perform distributed LLM post-training with SGLang?▼

Yes, you need Ray and PyTorch as core dependencies to perform distributed LLM post-training. The framework relies on Ray for distributed computing, PyTorch for model operations, and SGLang for high-throughput inference.

How does SGLang integrate with Megatron-LM for LLM post-training?▼

SGLang integrates with Megatron-LM by handling high-throughput inference while Megatron-LM manages distributed training. This tight integration allows efficient data generation and scaling for reinforcement learning tasks.

Related Skills

Explore More Software Engineering Skills→
zangjeicyzangjeicy

slime-rl-training

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

Community
Advanced
KarlinskySKarlinskyS

slime-rl-training

Coordinate Megatron-LM training with SGLang rollouts for RL-based post-training.

Community
Advanced
richardnguyen0715richardnguyen0715

slime-rl-training

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

Community
Advanced
evalstateevalstate

trl

Submit TRL training jobs on Hugging Face Jobs and convert models to GGUF.

Community
Advanced