skills.rest
Clay-HHK
Clay-HHKCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

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

2|Updated Apr 12, 2026
Search Tags:#llm#reinforcement-learning#rl#sglang#post-training#megatron-lm#data-buffer
One-click install
npx skills add https://github.com/Clay-HHK/claude-config --skill slime-rl-training-clay-hhk
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/Clay-HHK/claude-config/tree/main/skills/AI-research-SKILLs/06-post-training/slime
Command: npx skills add https://github.com/Clay-HHK/claude-config --skill slime-rl-training-clay-hhk

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

slime-rl-training provides a structured framework to guide LLM post-training with reinforcement learning using slime, unifying model training, high-throughput rollout generation, and data management to accelerate experiments and results.

Core Features & Use Cases

  • End-to-end RL post-training workflows for GLMs and other large language models
  • Megatron-LM-based training with SGLang rollout generation and a flexible data buffer
  • Multi-workflow support including standard GRPO, asynchronous, and multi-turn tool-enabled scenarios

Quick Start

Run a GRPO training workflow with slime using your dataset and a pre-configured model 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 run reinforcement learning post-training for LLMs using Megatron-LM and SGLang?▼

Reinforcement learning post-training for LLMs is guided by slime, which integrates Megatron-LM for training and SGLang for high-throughput rollout generation. You provide a dataset and pre-configured model script to run workflows like GRPO.

What RL algorithms are supported for LLM post-training?▼

LLM post-training supports configurable RL algorithms including grpo, gspo, ppo, and reinforce_plus_plus. These algorithms enable multi-turn and tool-use scenarios within the training workflow.

Does slime support multi-turn and tool-enabled scenarios in RL workflows?▼

Yes, slime supports multi-workflow scenarios including standard GRPO, asynchronous, and multi-turn tool-enabled scenarios. This allows for complex reinforcement learning post-training tasks.

How do I manage data generation and rollouts during LLM RL training?▼

LLM RL training manages data generation and rollouts using a flexible data buffer and SGLang. This unifies model training, high-throughput rollout generation, and data management to accelerate experiments.

Can I use slime for asynchronous RL training workflows?▼

Yes, you can use slime for asynchronous RL training workflows. It provides multi-workflow support including standard GRPO, asynchronous scenarios, and multi-turn tool-enabled tasks.

Do I need a specific data buffer setup to start scalable RL post-training?▼

Yes, scalable RL post-training requires a data buffer alongside Megatron-LM and SGLang integration. This setup enables custom data generation and high-throughput rollouts for LLMs.

Related Skills

Explore More Education & Research Skills→
ChimeraFoundationaChimeraFoundationa

slime-rl-training

Orchestrate RL-based post-training for large language models with Megatron-LM and SGLang rollout.

Community
Advanced
Lhhiep-maxcodeLhhiep-maxcode

nemotron-ultra

Retrieve NVIDIA Nemotron 3 Ultra architecture, training, evaluation, and release details from source files.

Community
Advanced
projectedanxprojectedanx
1

slime-rl-training

Train reasoning models with Megatron-LM and SGLang for RL post-training.

Community
Advanced
OpenLAIROpenLAIR
1.0k

verl-rl-training

Configure Verl backends and launch GRPO-based RL fine-tuning jobs.

Official
Advanced