skills.rest
adm-humanerd
adm-humanerdCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Integrate Megatron-LM with SGLang for end-to-end RL post-training of LLMs.

Updated Aug 27, 2026
Search Tags:#llm#reinforcement-learning#slime#rl#sglang#megatron-lm
One-click install
npx skills add https://github.com/adm-humanerd/drewgent --skill slime-rl-training-adm-humanerd
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/adm-humanerd/drewgent/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/adm-humanerd/drewgent --skill slime-rl-training-adm-humanerd

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

slime provides a structured framework for post-training RL on LLMs by integrating Megatron-LM training with SGLang through the slime workflow.

Core Features & Use Cases

  • End-to-end RL post-training for large language models using Megatron-LM and SGLang.
  • Data buffering and rollout generation to optimize training throughput across distributed GPUs.
  • Reward modeling and evaluation with pluggable reward functions and multi-turn tool use support.

Quick Start

Prepare your dataset and Megatron-LM configuration, then run the slime training script to begin 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 set up reinforcement learning post-training for LLMs using Megatron-LM and SGLang?▼

End-to-end RL post-training is achieved by interfacing Megatron-LM training with SGLang via the slime framework. You prepare your dataset and Megatron-LM configuration, then run the slime training script to start the process.

Can I use custom reward models for RL training with slime?▼

Yes, custom reward models are supported through optional adapters during RL training with slime. The framework allows pluggable reward functions and multi-turn tool use support to shape the training of your large language models effectively.

Does slime support multi-turn tool use during LLM rollout generation?▼

Multi-turn tool use is supported during LLM rollout generation in the slime framework. It provides modular data buffers and optional adapters to handle complex rollout generation and reward modeling across distributed GPUs.

How does weight synchronization work across distributed GPUs in slime RL training?▼

Weight synchronization across distributed GPUs is handled natively by the slime framework during RL training. It interfaces Megatron-LM training with SGLang to ensure consistent weight updates and optimized training throughput across all distributed nodes.

Is slime RL training applicable to models outside the GLM family?▼

Slime RL training is primarily applicable to GLM-family models and similar large language models. It covers data buffering, rollout generation, and reward shaping by integrating Megatron-LM with SGLang for distributed GPU training.

Related Skills

Explore More Software Engineering Skills→
ovachieverovachiever
52

grpo-rl-training

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

Community
Advanced
carterwayneskhizeinecarterwayneskhizeine
19

hermes-atropos-environments

Implements Hermes-AgentRL environments integrating multi-turn tool-calling loops with the Atropos training framework.

Community
Advanced
KarlinskySKarlinskyS

slime-rl-training

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

Community
Advanced
ovachieverovachiever
52

training-llms-megatron

Train large language models with Megatron-Core using configurable 3D and 4D parallelism.

Community
Advanced