slime-rl-training

Guide RL post-training of GLM models using slime with Megatron-LM and SGLang.

Updated Aug 27, 2026
One-click install
npx skills add https://github.com/zulumonkeymetallic/bob --skill slime-rl-training-zulumonkeymetallic
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/zulumonkeymetallic/bob/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/zulumonkeymetallic/bob --skill slime-rl-training-zulumonkeymetallic

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-rl-training provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.

Core Features & Use Cases

  • Guides on setting up RL post-training pipelines for LLMs using slime
  • Covers Megatron-LM integration and SGLang-driven rollout
  • Addresses data generation, buffering, and multi-turn RL scenarios

Quick Start

Follow the slime-rl-training guidance to configure an RL post-training workflow with slime.

Frequently Asked Questions about slime-rl-training

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

FAQPage Schema
How do I set up RL post-training for GLM models using slime and Megatron-LM?

slime provides a Megatron+SGLang framework for RL post-training of LLMs. You configure the pipeline using the slime-rl-training guidance to manage Megatron-LM integration, SGLang rollout, and data buffering for GLM models.

What is the slime framework used for in LLM post-training?

The slime framework provides a Megatron+SGLang architecture for scaling RL post-training of LLMs. It manages SGLang-driven rollout, data generation, and buffering workflows for GLM models.

Can I use slime for multi-turn RL scenarios and custom data generation?

Yes, slime explicitly addresses data generation, buffering, and multi-turn RL scenarios. The framework provides the necessary workflows to scale these reinforcement learning processes for GLM models.

Do I need SGLang and Ray installed to perform slime RL post-training?

Yes, slime requires sglang-router, Ray, torch, and transformers as dependencies. These libraries provide the necessary infrastructure for SGLang rollout and Megatron-LM integration during RL post-training.

How does SGLang rollout work with Megatron-LM in slime?

SGLang manages rollout generation while Megatron-LM handles training scaling within the slime framework. This integration enables efficient data generation and buffering workflows for RL post-training of GLM models.

What are the limitations of using slime for RL post-training?

slime is optimized for GLM models and requires tight Megatron-LM integration with SGLang. If your LLM architecture lacks GLM compatibility or your environment cannot support Ray and torch dependencies, slime may not suit your RL scaling needs.