skills.rest
AlexKoncept
AlexKonceptCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

slime-rl-training

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

Updated Aug 23, 2026
Search Tags:#llm#reinforcement-learning#slime#rl#sglang#post-training#megatron-lm
One-click install
npx skills add https://github.com/AlexKoncept/omnia-hub --skill slime-rl-training-alexkoncept
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/AlexKoncept/omnia-hub/tree/main/HERMES/optional-skills/mlops/slime
Command: npx skills add https://github.com/AlexKoncept/omnia-hub --skill slime-rl-training-alexkoncept

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Slime RL post-training guidance helps teams implement reinforcement-learning-based fine-tuning for large language models by unifying Megatron-LM and SGLang workflows.

Core Features & Use Cases

  • Training with Megatron-LM parallelism and RL optimization
  • Rollout generation via SGLang with data-buffering
  • Multi-turn/tool-enabled training and evaluation
  • Off-policy and async training options

Quick Start

Launch the slime RL training pipeline with your dataset and model configuration.

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 large language models?▼

Reinforcement learning post-training for large language models requires integrating Megatron-LM for parallel training with SGLang for rollout generation and data buffering to execute end-to-end fine-tuning workflows.

Can I perform asynchronous RL training and off-policy rollouts with SGLang?▼

SGLang supports both off-policy and asynchronous training options for LLM rollout generation, allowing flexible data buffering and multi-turn tool-enabled evaluation alongside standard Megatron-LM training modes.

Do I need Megatron-LM and SGLang configured before starting RL fine-tuning?▼

Megatron-LM and SGLang are required dependencies for RL fine-tuning, along with a configured Python environment and a data pipeline to support training execution and rollout generation.

How do I implement custom reward functions during LLM post-training?▼

RL post-training supports optional custom generate and reward functions, allowing researchers to tailor reinforcement learning optimization and data-generation workflows for GLM-family models and related architectures.

Does this RL training approach work with multi-turn and tool-enabled evaluation?▼

Multi-turn and tool-enabled training and evaluation are supported natively, enabling complex rollout generation via SGLang data buffering while maintaining Megatron-LM parallelism for large language models.

What is the best way to unify Megatron-LM training with SGLang rollout generation?▼

Unifying Megatron-LM training with SGLang rollout generation involves orchestrating data buffering between the two systems, enabling continuous RL optimization and generation cycles for fine-tuning GLM-family models.

Related Skills

Explore More Software Engineering Skills→
jason660519jason660519

slime-rl-training

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

Community
Advanced
OmidZamaniOmidZamani
120

dspy-finetune-bootstrap

Fine-tune a DSPy program into deployable model weights via BootstrapFinetune.

Community
Advanced
SriRamkunamsettySriRamkunamsetty

slime-rl-training

Orchestrate Megatron-LM and SGLang for end-to-end RL post-training.

Community
Advanced
ihatesea69ihatesea69
3

slime-rl-training

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

Community
Advanced