skills.rest
SethyPagna
SethyPagnaCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Automate reinforcement-learning post-training setup for large language models.

Updated May 14, 2026
Search Tags:#rl training#sglang#grpo#megatron-lm#reward modeling#llm post-training#ray orchestration
One-click install
npx skills add https://github.com/SethyPagna/Secretary-Jarvis --skill slime-rl-training-sethypagna
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/SethyPagna/Secretary-Jarvis/tree/main/src/capabilities/optional-skills/mlops/slime
Command: npx skills add https://github.com/SethyPagna/Secretary-Jarvis --skill slime-rl-training-sethypagna

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill removes the complexity of setting up reinforcement-learning post-training for large language models by turning model, data, rollout, and training requirements into a single coordinated workflow.

Core Features & Use Cases

  • Megatron-LM Training: Runs full-parallelism training for actor, critic, and other model components.
  • SGLang Rollouts: Generates high-throughput responses, rewards, and verifier outputs for RL loops.
  • Custom Workflows: Supports GRPO, PPO, async training, multi-turn tool use, and custom reward functions.
  • Use Case: Use it to prepare and operate a GLM, Qwen, DeepSeek, or Llama post-training run with the right data format, launch arguments, and troubleshooting guidance.

Quick Start

Ask me to set up a slime RL training workflow for your model, dataset, and GPU environment so I can produce the correct launch plan and configuration.

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 GRPO post-training for Llama or Qwen models?▼

GRPO post-training for Llama or Qwen models is set up by configuring Megatron-LM for full-parallelism training and SGLang for high-throughput rollouts. Ray orchestrates the workflow, coordinating data buffers and launch arguments for the RL loop.

Can I use Megatron-LM and SGLang for async rollout-training?▼

Yes, Megatron-LM and SGLang support async rollout-training. This Skill automates the Ray orchestration needed to manage actor and critic components alongside high-throughput response generation and verifier outputs.

What is the best way to configure custom reward pipelines for LLM post-training?▼

The best way to configure custom reward pipelines is to define the reward functions within your data format and let SGLang generate verifier outputs. Ray orchestration coordinates these rollouts with Megatron-LM training integration.

Does this RL training workflow support multi-turn agentic workflows?▼

Yes, this RL training workflow supports multi-turn agentic workflows. It handles async training and custom tool use by coordinating SGLang rollout serving with Megatron-LM training components.

Why do I need Ray orchestration for reinforcement-learning post-training?▼

Ray orchestration is needed to manage distributed GPU-aware configurations across Megatron-LM training and SGLang serving. It coordinates data-buffer handling and checkpoint management for large language models.

How do I troubleshoot GPU-aware configuration and checkpoint handling during PPO training?▼

Troubleshooting GPU-aware configuration and checkpoint handling during PPO training requires checking Ray orchestration logs and Megatron-LM parallelism settings. This Skill provides guidance for resolving launch argument and data format issues.

Related Skills

Explore More Software Engineering Skills→
ovachieverovachiever
52

grpo-rl-training

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

Community
Advanced
air-gappedair-gapped
5

vllm-quantization

Quantize LLM checkpoints for vLLM serving with FP8, NVFP4, and MXFP4 formats.

Official
Advanced
NITISH-gitbitNITISH-gitbit

slime-rl-training

Streamline Megatron-LM reinforcement learning post-training with SGLang rollout generation.

Community
Advanced
jason660519jason660519

slime-rl-training

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

Community
Advanced