skills.rest
zeronx798
zeronx798COMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Integrates Megatron-LM and SGLang for RL post-training of LLMs.

Updated Jul 13, 2026
Search Tags:#llm#reinforcement learning#sglang#grpo#post-training#megatron-lm
One-click install
npx skills add https://github.com/zeronx798/demo-hermes-agent --skill slime-rl-training-zeronx798
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/zeronx798/demo-hermes-agent/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/zeronx798/demo-hermes-agent --skill slime-rl-training-zeronx798

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sglang-router, ray, torch, transformers, and includes references (resource) components.

What problem does it solve?

This skill addresses the complexity of scaling Reinforcement Learning (RL) for large language models by providing a unified framework that bridges Megatron-LM training with high-throughput SGLang rollout generation.

Core Features & Use Cases

  • Hybrid RL Scaling: Integrates Megatron-LM for actor training with SGLang for efficient inference rollouts.
  • Flexible Training Workflows: Supports standard GRPO, asynchronous training, and multi-turn agentic training with custom tool-use logic.
  • Use Case: Use this skill when you need to perform post-training on large models like GLM-4 or Qwen3 using group-relative advantage estimation (GRPO) while maintaining high GPU utilization.

Quick Start

Execute the slime-rl-training skill to launch a GRPO training job using the qwen3-4B model configuration script.

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 post-training for large language models using Megatron-LM?▼

Scale LLM post-training with reinforcement learning by integrating Megatron-LM for distributed actor training with SGLang for high-throughput rollout generation. This hybrid approach bridges the two frameworks to maximize GPU utilization.

Can I use GRPO for post-training models like GLM-4 and Llama?▼

Yes, you can perform post-training on large models like GLM-4 and Llama using group-relative advantage estimation (GRPO). The framework supports standard GRPO workflows to optimize these architectures.

What is the best way to generate high-throughput inference rollouts during RL training?▼

Generate high-throughput inference rollouts during RL training by utilizing SGLang alongside Megatron-LM. This combination manages custom data buffers and facilitates efficient rollout generation for your models.

Does this reinforcement learning framework support multi-turn agentic training?▼

Yes, the framework supports multi-turn agentic training with custom tool-use logic. It facilitates asynchronous training and advantage estimation to satisfy complex agentic reasoning requirements.

Do I need Ray and PyTorch to run distributed GRPO training jobs?▼

Yes, Ray and PyTorch are required dependencies to run distributed GRPO training jobs. The framework relies on these alongside Transformers and the SGLang router to manage distributed training workloads.

Related Skills

Explore More Software Engineering Skills→
ldzhhxxldzhhxx
1

slime-rl-training

Run reinforcement learning post-training for LLMs with slime, Megatron-LM, and SGLang.

Community
Advanced
icyzhicyzh
1

slime-rl-training

Integrate Megatron-LM actor training with SGLang rollout generation for RL post-training.

Community
Advanced
Z43LZ43L

slime-rl-training

Integrates Megatron-LM training loops with accelerated SLang rollouts for LLM RL fine-tuning.

Community
Advanced
Tommy-ywTommy-yw
539

slime-rl-training

Optimize large language models with Megatron-LM training and SGLang rollouts.

Community
Advanced