skills.rest
matthew-johnson
matthew-johnsonCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

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

Updated Apr 20, 2026
Search Tags:#llm#reinforcement-learning#slime#rl-training#sglang#post-training#megatron-lm
One-click install
npx skills add https://github.com/matthew-johnson/hermes-agent --skill slime-rl-training-matthew-johnson
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/matthew-johnson/hermes-agent/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/matthew-johnson/hermes-agent --skill slime-rl-training-matthew-johnson

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sglang-router>=0.2.3, ray, torch>=2.0.0, transformers>=4.40.0, and includes references (resource) components.

What problem does it solve?

slime provides a structured framework to orchestrate reinforcement learning post-training for large language models, tying together Megatron-LM training, SGLang rollout generation, data buffering, and reward-based optimization to accelerate iterative improvement.

Core Features & Use Cases

  • End-to-end RL post-training workflows for LLMs using slime
  • Megatron-LM integration with SGLang-based rollout generation for high throughput
  • Custom data generation, reward modeling, and multi-turn agentic training for tool use and complex tasks
  • Use Case: Researchers can rapidly prototype RL objectives for GLM/Qwen/Llama models and evaluate improvements across iterations.

Quick Start

Run slime's RL post-training workflow by selecting a model, configuring rollout and reward settings, and launching the training 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 automate RL post-training workflows for large language models?▼

RL post-training workflows for large language models are automated using slime, which orchestrates Megatron-LM training, SGLang rollout generation, data buffering, and reward calculation into a unified training loop. It provides scripts to configure and launch iterative improvement cycles.

Does slime support Megatron-LM integration with SGLang for high-throughput rollout generation?▼

Yes, slime supports Megatron-LM integration paired with SGLang-based rollout generation to achieve high throughput. This combination accelerates iterative improvement by tying Megatron-LM training directly to fast rollout generation and data buffering.

Can I configure custom data generation and reward modeling for multi-turn agentic training?▼

Yes, you can configure custom data generation pipelines, reward modeling, and multi-turn agentic workflows for tool use. Slime provides modular configuration of training loops, data buffers, and reward calculation to handle complex tasks.

What do I need to set up before starting reinforcement learning post-training with slime?▼

You need Python dependencies including sglang-router, ray, torch, and transformers installed. You also need a selected model configuration, rollout settings, and reward calculation settings defined before launching the slime training script.

What's the best way to prototype RL objectives for Qwen or Llama models?▼

The best way to prototype RL objectives for GLM, Qwen, or Llama models is using slime's provided scripts and model configurations. Researchers can rapidly define reward settings, launch training loops, and evaluate improvements across iterations.

Why use slime for reinforcement learning scaling instead of other tools in the same category?▼

Slime is used for RL scaling because it provides a structured framework that ties together Megatron-LM training, SGLang rollout generation, and reward-based optimization. This integrated orchestration accelerates iterative improvement better than disparate tools.

Related Skills

Explore More Software Engineering Skills→
jonnabiojonnabio
6

slime-rl-training

Optimize LLMs post-training with RL using slime, Megatron-LM, and SGLang.

Community
Advanced
ovachieverovachiever
52

grpo-rl-training

Apply GRPO/RL training with TRL to optimize language models.

Community
Advanced
icyzhicyzh
1

slime-rl-training

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

Community
Advanced
LynxLabVNLynxLabVN

slime-rl-training

Automate post-training reinforcement learning optimization for large language models using the slime framework.

Official
Advanced