skills.rest
DaddyElonMusk69
DaddyElonMusk69COMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Scale RL-driven LLM fine-tuning with Megatron-LM and SGLang rollouts.

Updated Apr 12, 2026
Search Tags:#mlops#llm#fine-tuning#training#rl#sglang#megatron
One-click install
npx skills add https://github.com/DaddyElonMusk69/motis-agent --skill slime-rl-training-daddyelonmusk69
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/DaddyElonMusk69/motis-agent/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/DaddyElonMusk69/motis-agent --skill slime-rl-training-daddyelonmusk69

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill enables developers to perform reinforcement‑learning based fine‑tuning of large language models at scale, removing the complexity of manually coordinating Megatron‑LM training with high‑throughput SGLang rollouts.

Core Features & Use Cases

  • Integrated Training & Rollout: Seamlessly connects Megatron‑LM training loops with SGLang‑driven inference for RL algorithms such as GRPO, G‑SPO, and PPO.
  • Flexible Model Support: Works with GLM, Qwen, DeepSeek V3, Llama 3, and other Megatron‑compatible models.
  • Custom Data Pipelines: Provides buffered data sources, off‑policy replay, and multi‑turn agentic training workflows.
  • Use Case Example: A research team can launch a GRPO training run on a Qwen‑3‑4B model using their own JSONL dataset, monitor rewards via TensorBoard, and iterate rapidly without building custom infrastructure.

Quick Start

Run the slime skill to start a GRPO training job on your selected model and dataset.

Frequently Asked Questions about slime-rl-training

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

FAQPage Schema
How do I scale RL fine-tuning for large language models using Megatron-LM?▼

Scale RL fine-tuning by orchestrating Megatron-LM training loops with high-throughput SGLang inference rollouts, removing the need to manually coordinate distributed training and generation infrastructure.

Can I train Qwen or Llama 3 models with GRPO and custom JSONL datasets?▼

Yes, you can train Qwen, Llama 3, GLM, and DeepSeek V3 models using GRPO, G-SPO, or PPO algorithms with your own custom JSONL datasets and multi-turn agentic workflows.

What dependencies are needed for SGLang rollout and Megatron training?▼

SGLang rollout and Megatron training require sglang-router, Ray, PyTorch ≥ 2.0, and Transformers ≥ 4.40 to orchestrate the reinforcement learning pipeline.

Does slime support off-policy replay and buffered data sources for PPO training?▼

Yes, slime supports buffered data sources, off-policy replay, and multi-turn agentic training workflows for PPO, GRPO, and G-SPO algorithms.

Why use SGLang instead of standard inference for RL LLM training?▼

SGLang provides high-throughput inference rollouts specifically designed to integrate seamlessly with Megatron-LM training loops, enabling rapid iteration on reward signals without building custom infrastructure.

How to monitor rewards during GRPO training on a Qwen-3-4B model?▼

Monitor rewards during GRPO training by integrating TensorBoard, allowing you to track reward metrics and iterate rapidly on your Qwen-3-4B model without building custom monitoring infrastructure.

Related Skills

Explore More Software Engineering Skills→
zangjeicyzangjeicy

slime-rl-training

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

Community
Advanced
vasilyu1983vasilyu1983
73

ai-llm-inference

Optimize LLM inference performance and cost with vLLM and quantization.

Community
Advanced
blueskies1818blueskies1818
1

slime-rl-training

Guide LLM post-training with RL using Megatron-LM and SGLang.

Community
Advanced
nadicodeainadicodeai

slime-rl-training

Configure RL fine-tuning pipelines for GLM models with Megatron-LM and SGLang.

Official
Advanced