skills.rest
BermudaLocals
BermudaLocalsCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

slime-rl-training

Automate RL-based post-training workflows for large language models with Megatron-LM and SGLang.

1|1|Updated Apr 26, 2026
Search Tags:#llm#reinforcement-learning#slime#rl#sglang#post-training#megatron-lm
One-click install
npx skills add https://github.com/BermudaLocals/hermes-agent-lite --skill slime-rl-training-bermudalocals
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/BermudaLocals/hermes-agent-lite/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/BermudaLocals/hermes-agent-lite --skill slime-rl-training-bermudalocals

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?

Provides a guided, scalable framework for RL-based post-training of large language models by integrating Megatron-LM with SGLang, enabling efficient rollout, training, and model alignment workflows.

Core Features & Use Cases

  • End-to-end RL post-training workflows for GLMs and transformer-based models using Megatron-LM + SGLang.
  • Support for standard GRPO-style training, asynchronous update modes, and multi-turn / tool-use scenarios.
  • Robust data buffering, configurable model/scripts, and multi-argument coordination to manage rollout, reward computation, and weight synchronization.

Quick Start

Run slime to begin RL-based post-training for your GLM models with Megatron-LM and SGLang.

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 reinforcement learning post-training for my LLM?▼

RL-based post-training workflows are automated by integrating Megatron-LM with SGLang to handle scalable rollout, training, and model alignment. This framework supports configurable data buffers and model scripts to manage reward computation and weight synchronization.

What is the best way to run GRPO-style training on transformer models?▼

Running GRPO-style training is best achieved by using slime to coordinate Megatron-LM and SGLang for scalable rollout and training. This approach supports standard workflows, asynchronous updates, and multi-turn tool-use scenarios.

Can I use SGLang with Megatron-LM for multi-turn tool-use scenarios?▼

Yes, SGLang works with Megatron-LM to support multi-turn and tool-use scenarios during RL post-training. The integration manages multi-argument coordination to handle rollout, reward computation, and weight synchronization.

Do I need Ray and PyTorch for asynchronous RL updates?▼

Yes, Ray and PyTorch are required dependencies for executing asynchronous RL updates and managing scalable rollout operations. They provide the distributed execution and tensor computation foundation for transformer model alignment.

Does slime support asynchronous updates for transformer-based language models?▼

Yes, slime supports asynchronous update modes for GLMs and transformer-based models during RL post-training. It coordinates SGLang and Megatron-LM to manage rollout, reward computation, and weight synchronization.

Related Skills

Explore More Education & Research Skills→
zangjeicyzangjeicy

slime-rl-training

Train LLMs with GRPO and PPO using Megatron-LM and SGLang.

Community
Advanced
jacardljacardl

slime-rl-training

Run RL post-training for large language models with Megatron-LM and SGLang.

Community
Advanced
ZardLi1115ZardLi1115
4

slime-rl-training

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

Community
Advanced
qcmuuqcmuu

verl-rl-training

Orchestrate verl-based RL training loops for LLM post-training.

Community
Advanced