skills.rest
gagan114662
gagan114662COMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Integrates Megatron-LM and SGLang for RL-based LLM post-training with flexible data buffering.

Updated Jan 30, 2026
Search Tags:#llm#reinforcement learning#glm#sglang#rlhf#post-training#megatron-lm
One-click install
npx skills add https://github.com/gagan114662/content_books --skill slime-rl-training-gagan114662
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/gagan114662/content_books/tree/main/AI-research-SKILLs/06-post-training/slime
Command: npx skills add https://github.com/gagan114662/content_books --skill slime-rl-training-gagan114662

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 designed for scaling training with frameworks like Megatron-LM and SGLang.

Core Features & Use Cases

  • Integrated Training & Inference: Seamlessly combines Megatron-LM for efficient training with SGLang for high-throughput inference rollouts.
  • Flexible Data Workflows: Supports custom data generation and management through a flexible data buffer system.
  • Model Support: Compatible with a wide range of models including GLM, Qwen, DeepSeek, and Llama.
  • Use Case: Train a new version of a GLM model using Reinforcement Learning from Human Feedback (RLHF) by leveraging custom data generation and the framework's efficient rollout mechanism.

Quick Start

Use the slime-rl-training skill to launch GRPO training with the qwen3-4B model using the provided training data.

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 LLM RLHF training with Megatron-LM and SGLang?▼

You can scale LLM RLHF training by integrating Megatron-LM for efficient training and SGLang for high-throughput inference rollouts. This combination supports asynchronous training and multi-turn agentic workflows.

What is the best way to run reinforcement learning post-training on a Qwen model?▼

Reinforcement learning post-training on a Qwen model is best handled by using an integrated framework that combines high-throughput inference with flexible data buffering. This streamlines the rollout mechanism for custom data generation.

Does SGLang work with Megatron-LM for asynchronous LLM training?▼

Yes, SGLang works with Megatron-LM to facilitate asynchronous LLM training. This integration enables advanced training workflows including multi-turn agentic training and efficient rollouts.

Can I use GLM, DeepSeek, and Llama architectures for large-scale LLM post-training?▼

Yes, you can use GLM, DeepSeek, and Llama architectures for large-scale LLM post-training. The framework supports a wide range of models for custom data generation and reinforcement learning.

How does custom data generation work during LLM reinforcement learning?▼

Custom data generation during LLM reinforcement learning works through a flexible data buffer system. This allows you to manage and generate training data efficiently while leveraging inference rollouts.

What dependencies do I need to run GRPO training with Megatron-LM?▼

To run GRPO training with Megatron-LM, you need dependencies including sglang-router, ray, torch, and transformers. These frameworks enable the distributed training and high-throughput inference required.

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
superfhpsuperfhp

serving-llms-vllm

Serve LLMs with OpenAI-compatible endpoints using vLLM.

Community
Advanced
GarrettRoiGarrettRoi

slime-rl-training

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

Community
Advanced
shuff57shuff57

fine-tuning-with-trl

Align language models with human preferences using Transformer Reinforcement Learning methods.

Community
Advanced