skills.rest
kotakbiasa
kotakbiasaCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

slime-rl-training

Coordinate RL-based post-training for large language models with Megatron-LM and SGLang.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

slime provides a guided framework for RL-based post-training of large language models by integrating Megatron-LM training with SGLang rollout, data buffers, and memory management to streamline iterative improvement.

Core Features & Use Cases

  • End-to-end RL post-training workflows for GLMs/LLMs with scalable rollout
  • Flexible data generation, reward modeling, and custom generate/reward functions
  • Off-policy and async training configurations with model scripting support

Quick Start

Provide a minimal RL post-training setup by configuring slime to run GRPO-based rollout and Megatron-LM training with 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 run RL post-training for large language models using Megatron-LM and SGLang?▼

RL post-training for large language models is executed by integrating Megatron-LM for training and SGLang for rollout. This framework coordinates the end-to-end loop, spanning rollout, data buffers, and reward modeling to streamline iterative improvement.

Can I use custom reward functions during LLM reinforcement learning rollout?▼

Custom reward functions are fully supported during LLM reinforcement learning rollout. The framework allows flexible data generation and reward modeling, enabling you to inject custom logic directly into the training and evaluation loops.

What is the best way to scale GRPO-based rollout for GLM post-training in distributed GPU environments?▼

Scaling GRPO-based rollout for GLM post-training in distributed GPU environments requires coordinating Megatron-LM training with SGLang. This setup manages memory buffering and off-policy configurations to handle scalable, asynchronous workloads.

Does slime support asynchronous and off-policy training configurations?▼

Asynchronous and off-policy training configurations are supported by slime. It manages memory buffering and generation workflows to enable flexible, scalable training loops without requiring strict on-policy data synchronization.

What arguments do I need to configure for end-to-end LLM reinforcement learning workflows?▼

End-to-end LLM reinforcement learning workflows require configurations across Megatron-LM, SGLang, and slime argument categories. You must define parameters for rollout generation, memory buffers, and reward modeling to execute the training loop.

Why do I need SGLang for rollout in LLM reinforcement learning?▼

SGLang is needed for rollout in LLM reinforcement learning to handle scalable generation and memory management. It pairs with Megatron-LM training to coordinate the iterative data generation and model updating loop efficiently.

Related Skills

Explore More Software Engineering Skills→
JuanMS20JuanMS20

huggingface-accelerate

Configure distributed PyTorch training across multi-GPU and multi-node devices.

Community
Advanced
markwang2658markwang2658
31

slime-rl-training

Plan and execute LLM RL post-training with the slime Megatron+SGLang framework.

Community
Advanced
ovachieverovachiever
52

grpo-rl-training

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

Community
Advanced
Quill-AgentQuill-Agent
3

slime-rl-training

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

Community
Advanced