skills.rest
GregsGreyCode
GregsGreyCodeCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Coordinate end-to-end RL training for LLMs with Megatron and SGLang.

6|2|Updated Mar 19, 2026
Search Tags:#distributed-training#reinforcement-learning#rl#sglang#llm-training#megatron-lm#data-buffer
One-click install
npx skills add https://github.com/GregsGreyCode/Logos --skill slime-rl-training-gregsgreycode
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/GregsGreyCode/Logos/tree/main/skills/mlops/training/slime
Command: npx skills add https://github.com/GregsGreyCode/Logos --skill slime-rl-training-gregsgreycode

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Coordinating end-to-end RL training for large language models (LLMs) using slime's Megatron+SGLang framework, reducing integration overhead and enabling scalable experiments.

Core Features & Use Cases

  • LLM post-training with RL using slime, compatible with GLM family, Llama 3, and Qwen-like models.
  • Data generation workflows, flexible buffers, and multi-turn or tool-assisted training scenarios.
  • Production-grade orchestration with modular configuration and experiment tracking for reproducibility.

Quick Start

Run the training workflow with your model, data, and slime configuration to start RL training.

Frequently Asked Questions about slime-rl-training

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

FAQPage Schema
How do I coordinate end-to-end RL training for LLMs using Megatron-LM and SGLang?▼

To coordinate end-to-end RL training for LLMs, you can use slime's Megatron+SGLang framework to manage Megatron-LM training and SGLang-backed rollout. This reduces integration overhead and enables scalable, production-grade experiments.

Can I use distributed RL training for multi-turn or tool-assisted scenarios?▼

Yes, distributed RL training supports multi-turn and tool-assisted scenarios. The framework allows you to apply custom generate and reward functions to handle these complex, interactive training workflows effectively.

What models are compatible with slime for reinforcement learning post-training?▼

Reinforcement learning post-training with slime is compatible with GLM family, Llama 3, and Qwen-like models. This allows you to apply scalable RL techniques across various popular open-source large language model architectures.

How do I configure modular arguments for Megatron, SGLang, and slime?▼

You configure modular arguments by separating parameters into three distinct categories: Megatron, SGLang, and slime. This modular configuration enforces robust experiment tracking and reproducibility across your distributed training workflows.

Does slime support data buffers for LLM training workflows?▼

Yes, slime supports data buffers for LLM training workflows. These flexible buffers integrate with data generation workflows to manage and streamline the inputs required for scalable reinforcement learning experiments.

Related Skills

Explore More Software Engineering Skills→
ovachieverovachiever
52

grpo-rl-training

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

Community
Advanced
ScientiaCapitalScientiaCapital
1

model-deployment

Export fine-tuned models to GGUF and deploy with vLLM or Ollama.

Community
Advanced
Yangel-hideYangel-hide

slime-rl-training

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

Community
Advanced
evalstateevalstate

trl

Submit TRL training jobs on Hugging Face Jobs and convert models to GGUF.

Community
Advanced