skills.rest
heysuhas
heysuhasCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

slime-rl-training

Integrate Megatron-LM and SGLang for large-scale LLM reinforcement learning training.

2|1|Updated Jul 14, 2026
Search Tags:#llm#reinforcement learning#sglang#grpo#post-training#megatron-lm
One-click install
npx skills add https://github.com/heysuhas/hermes_cli --skill slime-rl-training-heysuhas
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/heysuhas/hermes_cli/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/heysuhas/hermes_cli --skill slime-rl-training-heysuhas

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sglang-router, ray, torch, transformers, and includes references (resource) components.

What problem does it solve?

This skill addresses the complexity of scaling Reinforcement Learning (RL) for large language models by integrating Megatron-LM for training with SGLang for high-throughput rollout generation.

Core Features & Use Cases

  • Megatron-LM Integration: Enables native training with full parallelism (TP, PP, DP, SP) for massive models.
  • High-Throughput Rollout: Utilizes SGLang for efficient response generation and reward verification.
  • Flexible Workflows: Supports standard GRPO, asynchronous training, and multi-turn agentic training for complex reasoning tasks.

Quick Start

Use the slime-rl-training skill to launch a GRPO training job for a Qwen3-4B model by sourcing the model configuration script and executing the training command with your prompt data.

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 reinforcement learning post-training for large language models?▼

Scale LLM post-training using reinforcement learning by integrating Megatron-LM for distributed training with SGLang for high-throughput rollout generation. This combination supports full parallelism and efficient reward verification across massive models.

Can I use GRPO and asynchronous training loops for LLM post-training?▼

Yes, GRPO and asynchronous training loops are fully supported for LLM post-training. The framework enables flexible workflows including standard GRPO, asynchronous training, and multi-turn agentic training for complex reasoning tasks.

Does Megatron-LM integration support tensor and pipeline parallelism for RL training?▼

Megatron-LM integration supports native training with full parallelism for RL training. It enables tensor, pipeline, data, and sequence parallelism to efficiently train massive language models during the reinforcement learning phase.

What distributed environment do I need for high-throughput rollout generation?▼

High-throughput rollout generation requires a distributed environment with Ray, PyTorch, and SGLang. These dependencies manage high-throughput response generation and model weight synchronization across the distributed training cluster.

How do I launch a GRPO training job for a Qwen model?▼

Launch a GRPO training job for a Qwen model by sourcing the model configuration script and executing the training command with your prompt data. The framework supports models like GLM, Qwen, and Llama.

Related Skills

Explore More Software Engineering Skills→
zangjeicyzangjeicy

slime-rl-training

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

Community
Advanced
Chris-Chai-MinjaeChris-Chai-Minjae

slime-rl-training

Guide RL post-training for large language models with slime.

Community
Advanced
hhhi21ghhhi21g

slime-rl-training

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

Community
Advanced
ricablericable

slime-rl-training

Orchestrate RL post-training for large language models with slime.

Community
Advanced