skills.rest
tadod12
tadod12COMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Configure Megatron-LM and SGLang arguments for post-training RL workflows.

Updated Mar 18, 2026
Search Tags:#llm#experimentation#reinforcement-learning#slime#rl#post-training#megatron-sglang
One-click install
npx skills add https://github.com/tadod12/fraud-detection-research --skill slime-rl-training-tadod12
Or copy as Structured Prompt for Agentâ–¼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/tadod12/fraud-detection-research/tree/main/.agent/skills/06-post-training/slime
Command: npx skills add https://github.com/tadod12/fraud-detection-research --skill slime-rl-training-tadod12

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Guides researchers to perform post-training RL on large language models 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

  • End-to-end RL workflows for LLMs with slime, including Megatron-LM integration, SGLang-based rollout, data buffering, and evaluation.
  • Flexible experiment design for GLM-family models with custom data generation and multi-turn interactions.
  • Reproducible research pipelines with configurable RL algorithms (grpo, gspo, ppo, reinforce_plus_plus) and off-policy training options.

Quick Start

Run the slime RL-training workflow with your GLM model and dataset to begin post-training reinforcement learning experiments.

Frequently Asked Questions about slime-rl-training

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I perform reinforcement learning post-training on LLMs using Megatron-LM and SGLang?â–¼

Reinforcement learning post-training on LLMs is performed by configuring Megatron-LM arguments, enabling high-throughput rollout generation via SGLang, and setting up data buffers for scalable RL workflows.

Can I use slime to train GLM-family models with custom data generation workflows?â–¼

Yes, slime supports training GLM-family models with custom data generation workflows, allowing you to implement custom generate functions and handle multi-turn interactions for flexible experiment design.

What RL algorithms are available for large language model post-training in slime?â–¼

Slime supports configurable RL algorithms including grpo, gspo, ppo, and reinforce_plus_plus, providing reproducible research pipelines with off-policy training options for large language model post-training.

How do I set up end-to-end off-policy training pipelines for LLMs?â–¼

End-to-end off-policy training pipelines are configured by enforcing Megatron-LM and SGLang arguments, managing data buffers, and establishing evaluation workflows for reproducible reinforcement learning experiments.

Does slime support high-throughput rollout generation for scalable RL workflows?â–¼

Yes, slime enables high-throughput rollout generation via SGLang integration, specifically designed to scale reinforcement learning workflows for large language models requiring tight Megatron-LM integration.

Related Skills

Explore More Education & Research Skills→
dawsonblockdawsonblock

slime-rl-training

Implement RL post-training for GLM models with slime, Megatron-LM, and SGLang.

Community
Advanced
sencersoylusencersoylu

computational-study

Plan and execute computational experiments for algorithm benchmarking and ablation studies.

Community
Advanced
ruiyangruiyiruiyangruiyi

slime-rl-training

Orchestrates Megablock-LM RL training with SGLab-based rollout generation for GRPO-style workflows.

Community
Advanced
zhouboyu-xrealzhouboyu-xreal

slime-rl-training

Scale reinforcement learning post-training with Megatron-LM and SGLang.

Community
Advanced