skills.rest
Yangel-hide
Yangel-hideCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Guide RL post-training of large language models with Megatron-LM and SGLang.

Updated May 3, 2026
Search Tags:#llm#training#reinforcement-learning#slime#rl#sglang#megatron-lm
One-click install
npx skills add https://github.com/Yangel-hide/video-production-planner-agent --skill slime-rl-training-yangel-hide
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/Yangel-hide/video-production-planner-agent/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/Yangel-hide/video-production-planner-agent --skill slime-rl-training-yangel-hide

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

slime guides RL post-training of LLMs using slime framework, enabling end-to-end RL workflows with Megatron-LM and SGLang.

Core Features & Use Cases

  • Integrates Megatron-LM with SGLang rollout for scalable RL post-training.
  • Provides data-buffering, multi-turn tool interactions, and custom generate/reward flows for GLM-family models.
  • Use cases include RLHF-style fine-tuning, policy optimization, and iterative evaluation of RL agents.

Quick Start

Install slime, configure Megatron-SGLang settings, and begin an RL post-training workflow for your GLM/Large LM.

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 large language models using Megatron-LM and SGLang?▼

To set up RL post-training for large language models, install the slime framework, configure your Megatron-LM and SGLang rollout settings, and initialize the training workflow. This enables scalable reinforcement learning directly on your GLM-family models.

Can I implement multi-turn tool interactions and data buffering during LLM reinforcement learning?▼

Yes, you can implement multi-turn tool interactions and data buffering during LLM reinforcement learning. The slime framework natively supports these workflows, allowing complex, iterative policy optimization and evaluation for large language models.

How do I configure custom generate and reward functions for GRPO workflows?▼

You can configure custom generate and reward functions for GRPO workflows by defining them within the slime framework's training configuration. This allows tailored policy optimization and specific reward signals during reinforcement learning.

Does slime support asynchronous RL training workflows for LLMs?▼

Yes, slime supports asynchronous RL training workflows for LLMs. The framework provides specific configurations for async operations, enabling efficient rollout and policy optimization alongside Megatron-LM and SGLang integration.

What is the best way to troubleshoot rollout or training issues in an RL post-training pipeline?▼

The best way to troubleshoot RL post-training issues is to use the slime framework's built-in troubleshooting guide. It addresses common problems in Megatron-LM integration, SGLang rollout, and data buffering to resolve pipeline failures.

How does integrating SGLang with Megatron-LM improve reinforcement learning for LLMs?▼

Integrating SGLang with Megatron-LM improves reinforcement learning by enabling scalable, efficient rollout generation. This combination within the slime framework optimizes end-to-end RL post-training workflows for large language models.

Related Skills

Explore More Software Engineering Skills→
ovachieverovachiever
52

training-llms-megatron

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

Community
Advanced
tadod12tadod12

slime-rl-training

Configure Megatron-LM and SGLang arguments for post-training RL workflows.

Community
Advanced
eddieluengeddielueng
1

slime-rl-training

Run GRPO RL post-training for LLMs with Megatron-LM and SGLang rollouts.

Community
Advanced
gagan114662gagan114662

fine-tuning-with-trl

Fine-tune LLMs with TRL using SFT, DPO, PPO, and GRPO.

Community
Advanced