skills.rest
wwwillott
wwwillottCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Automate scalable RL post-training for LLMs with Megatron-LM and SGLang.

Updated Aug 23, 2026
Search Tags:#llm#reinforcement-learning#rl-training#sglang#post-training#megatron-lm
One-click install
npx skills add https://github.com/wwwillott/jobnimbus --skill slime-rl-training-wwwillott
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/wwwillott/jobnimbus/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/wwwillott/jobnimbus --skill slime-rl-training-wwwillott

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

slime provides a scalable post-training framework for large language models by integrating Megatron-LM training with SGLang-powered rollout generation, enabling efficient RL-based fine-tuning at scale.

Core Features & Use Cases

  • End-to-end RL post-training for GLM-family models (GLM-4.x, Qwen3, DeepSeek, Llama3) with high-throughput rollout and on-the-fly evaluation.
  • Modular data handling with a flexible data buffer and pre-configured model scripts to accelerate experimentation.
  • Production-grade workflow support for multi-GPU, distributed training, and customizable reward models.

Quick Start

Source a preconfigured model script and start the RL post-training workflow using the provided train.py command.

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 post-training for large language models across multiple GPUs?▼

Scale RL post-training for large language models by integrating Megatron-LM training with SGLang rollout generation. This framework automates end-to-end distributed workflows across multi-GPU environments for efficient fine-tuning.

Can I use slime for RL fine-tuning on Qwen3 and Llama3 models?▼

Yes, you can perform RL fine-tuning on Qwen3 and Llama3 models. The framework provides pre-configured model scripts targeting end-to-end workflows for GLM-4.x, DeepSeek, Qwen3, and Llama3 architectures.

How does SGLang rollout generation work with Megatron-LM training?▼

SGLang handles high-throughput rollout generation to produce training data, while Megatron-LM executes the distributed training updates. This modular integration automates efficient RL post-training for large language models.

What is the best way to manage data buffers for LLM reinforcement learning?▼

Manage data buffers for LLM reinforcement learning using the framework's modular data handling system. It features a flexible data buffer and pre-configured scripts to accelerate experimentation and streamline data management.

How do I start an RL training workflow with customizable reward models?▼

Start an RL training workflow with customizable reward models by sourcing a preconfigured model script and executing the provided train.py command. This initiates the automated end-to-end training process.

Do I need a multi-GPU environment for LLM post-training with slime?▼

Yes, a multi-GPU environment is required for LLM post-training. The framework satisfies requirements for distributed training and production-grade workflows, utilizing Megatron-LM to handle scalable reinforcement learning.

Related Skills

Explore More Software Engineering Skills→
CUexterCUexter
2

slime-rl-training

Run RL-based LLM post-training with Megatron-LM and SGLang for scalable rollout and GRPO workflows.

Community
Advanced
Jonah-1Jonah-1

eval

Run inference on neural network checkpoints and generate PSNR, SSIM, and LPIPS metrics.

Community
Intermediate
matthew-johnsonmatthew-johnson

slime-rl-training

Automate RL post-training workflows for large language models with slime.

Community
Advanced
jason660519jason660519

slime-rl-training

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

Community
Advanced