skills.rest
Tommy-yw
Tommy-ywCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

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

539|39|Updated May 1, 2026
Search Tags:#optimization#LLM#training#RL#SGLang#Megatron-LM
One-click install
npx skills add https://github.com/Tommy-yw/RunbookHermes --skill slime-rl-training-tommy-yw
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/Tommy-yw/RunbookHermes/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/Tommy-yw/RunbookHermes --skill slime-rl-training-tommy-yw

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) and assets (resource) components.

What problem does it solve?

This Skill addresses the challenge of optimizing large language models (LLMs) for reinforcement learning (RL) post-training, providing a scalable solution for production environments.

Core Features & Use Cases

  • Megatron-LM Training: Full parallelism support with Megatron-LM for actor model training.
  • SGLang Rollout: High-throughput rollout generation using SGLang and router.
  • Data Buffer Management: Flexible prompt management and sample storage.
  • Model Compatibility: Supports GLM, Qwen, DeepSeek, Llama, and more.
  • Use Case: Ideal for research and production scenarios requiring tight integration of Megatron-LM with SGLang for RL scaling.

Quick Start

Start the slime skill to train a GRPO model with custom configurations.

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?▼

To scale LLM reinforcement learning post-training, you can use the slime framework to connect Megatron-LM for distributed actor training with SGLang for high-throughput rollout generation in production environments.

What models are compatible with Megatron-LM and SGLang for RL training?▼

The RL training setup supports model compatibility for GLM, Qwen, DeepSeek, and Llama architectures, integrating Megatron-LM for training and SGLang for rollout generation.

How do I generate GRPO rollouts with high throughput using SGLang?▼

You can generate GRPO rollouts with high throughput by using SGLang and its router, which manages prompt generation and sample storage within the slime RL training workflow.

Can I use Megatron-LM for full parallelism during LLM RL post-training?▼

Yes, Megatron-LM provides full parallelism support for actor model training during LLM RL post-training, enabling scalable optimization for production environments.

What dependencies are required to start RL training with Megatron-LM and SGLang?▼

You need sglang-router, ray, torch, and transformers to run the RL training framework, which integrates Megatron-LM and SGLang for scalable model optimization.

Is slime-rl training framework suitable for production environments?▼

Yes, the slime RL training framework is designed for production environments, offering custom data generation workflows and tight Megatron-LM integration for scaling LLM optimization.

Related Skills

Explore More Software Engineering Skills→
tianhao909tianhao909
1

slime-rl-training

Integrates Megatron-LM and SGLang for scalable GRPO-based LLM reinforcement post-training.

Community
Advanced
jayll1303jayll1303
17

tensorrt-llm

Convert HuggingFace checkpoints into optimized TensorRT-LLM engines with quantization.

Community
Advanced
ovachieverovachiever
52

training-llms-megatron

Train large language models with Megatron-Core using configurable 3D and 4D parallelism.

Community
Advanced
hhhi21ghhhi21g

slime-rl-training

Scale RL post-training for LLMs with Megatron-LM and SGLang.

Community
Advanced