skills.rest
KappTech88
KappTech88COMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Deploy RLHF-style post-training pipelines for LLMs with Megatron-LM and SGLang.

Updated Mar 30, 2026
Search Tags:#rollout#reinforcement-learning#sglang#llm-training#rlhf#megatron-lm#slime-framework
One-click install
npx skills add https://github.com/KappTech88/AI-RESEARCH-SKILLS-MCP --skill slime-rl-training-kapptech88
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/KappTech88/AI-RESEARCH-SKILLS-MCP/tree/main/skills/slime
Command: npx skills add https://github.com/KappTech88/AI-RESEARCH-SKILLS-MCP --skill slime-rl-training-kapptech88

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

slime-rl-training provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.

Core Features & Use Cases

  • Megatron-LM based training with SGLang rollout for high-throughput data generation
  • Data buffering and multi-turn reasoning workflows for tool use
  • Support for GLM family models (GLM-4.x, Qwen3, DeepSeek V3/R1, Llama 3) with configuration examples
  • Production-ready workflow guardrails and troubleshooting guidance (see references)

Quick Start

Launch a slime RL post-training run with your chosen model and dataset.

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 RLHF post-training for large language models using Megatron-LM and SGLang?▼

Scale RLHF post-training by deploying the slime framework to orchestrate Megatron-LM training with SGLang rollout. This setup enables high-throughput data generation, model parallelism, and multi-turn reasoning workflows for large language models.

What is the best way to implement high-throughput rollout for RLHF data generation?▼

High-throughput rollout for RLHF data generation is implemented using SGLang integrated within the slime framework. This configuration supports efficient data buffering and generation required for scaling reinforcement learning pipelines.

Can I use the slime framework for RL training on GLM-4 or Qwen3 models?▼

Yes, the slime framework supports RL training on GLM family models, including GLM-4.x, Qwen3, DeepSeek V3/R1, and Llama 3. It provides specific configuration examples to deploy end-to-end RLHF pipelines for these architectures.

Does slime support custom data generation workflows and multi-turn reasoning for tool use?▼

Yes, slime supports custom data generation workflows and multi-turn reasoning for tool use. The framework applies data buffering and SGLang rollout to manage complex, multi-turn reasoning workflows during RL post-training.

What are the limitations of using slime for reinforcement learning post-training?▼

Limitations include the strict dependency on Megatron-LM and SGLang setups for model parallelism and rollout. Users must configure specific dependencies and workflow requirements, which may restrict flexibility outside this specific tech stack.

Related Skills

Explore More Education & Research Skills→
ovachieverovachiever
52

grpo-rl-training

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

Community
Advanced
BirgerMoellBirgerMoell

megatron-hf-converter — Claude Code skill guide

Convert HuggingFace and Megatron-LM checkpoints with numerical equivalence validation.

Community
Intermediate
matthew-johnsonmatthew-johnson

slime-rl-training

Automate RL post-training workflows for large language models with slime.

Community
Advanced
tianhao909tianhao909
1

slime-rl-training

Integrates Megatron-LM and SGLang for scalable GRPO-based LLM reinforcement post-training.

Community
Advanced