skills.rest
AissenLiu
AissenLiuCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

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

Updated Apr 27, 2026
Search Tags:#multi-turn#reinforcement-learning#slime#rl-training#sglang#megatron-lm#data-buffer
One-click install
npx skills add https://github.com/AissenLiu/EasyHermes --skill slime-rl-training-aissenliu
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/AissenLiu/EasyHermes/tree/main/hermes-agent/optional-skills/mlops/slime
Command: npx skills add https://github.com/AissenLiu/EasyHermes --skill slime-rl-training-aissenliu

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sglang-router>=0.2.3, ray, torch>=2.0.0, transformers>=4.40.0, and includes references (resource) components.

What problem does it solve?

slime provides an end-to-end RL post-training workflow by integrating Megatron-LM training with SGLang rollout, enabling scalable experiments without bespoke glue code.

Core Features & Use Cases

  • End-to-end RL post-training for large language models (GLM, Qwen, DeepSeek, Llama)
  • Integrated data buffering, multi-turn tool-enabled interactions, and custom reward mechanics
  • Async training options and evaluation workflows for research and production pilots

Quick Start

Launch a slime training workflow by selecting a pre-configured model script and supplying your dataset path to train with Megatron-LM and SGLang.

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 end-to-end RL post-training for large language models using Megatron-LM and SGLang?▼

You can perform end-to-end RL post-training by selecting a pre-configured model script and supplying your dataset path to integrate Megatron-LM training with SGLang rollout. The system handles data buffering and orchestration for scalable experiments without requiring bespoke glue code.

Does RL training with slime support multi-turn agentic interactions and custom reward functions?▼

Yes, RL training with slime supports multi-turn agentic interactions and custom reward functions. The system provides integrated data buffering and customizable model scripts to handle complex multi-turn tool-enabled interactions during training.

Can I use this workflow to train models like Qwen, DeepSeek, and Llama asynchronously?▼

Yes, you can train Qwen, DeepSeek, and Llama models asynchronously. The workflow provides async training options alongside standard training and evaluation workflows for research and production pilots.

What dependencies are required to run slime RL training with Megatron-SGLang?▼

You need sglang-router, Ray, PyTorch, and Transformers to run slime RL training with Megatron-SGLang. These dependencies provide the distributed infrastructure and model handling capabilities required for the workflow.

How does the data buffer orchestration work during multi-turn tool-enabled interactions?▼

Data buffer orchestration coordinates the data pipeline between Megatron-LM training and SGLang rollout during multi-turn tool-enabled interactions. It structures the configuration system to ensure reproducible and scalable RL training experiments.

Related Skills

Explore More Software Engineering Skills→
richardnguyen0715richardnguyen0715

slime-rl-training

Train LLMs with reinforcement learning using Megatron-LM and SGLang.

Community
Advanced
automatedigitalautomatedigital
1

unsloth

Fine-tune large language models with LoRA and QLoRA optimization.

Official
Intermediate
KarlinskySKarlinskyS

slime-rl-training

Coordinate Megatron-LM training with SGLang rollouts for RL-based post-training.

Community
Advanced
davpatel605-beepdavpatel605-beep

slime-rl-training

Configure scalable RL post-training for LLMs using the slime framework.

Community
Advanced