skills.rest
nelohenriq
nelohenriqCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Orchestrates post-training reinforcement learning workflows for LLMs using slime, Megatron-LM integration, and SGLang rollout.

1|Updated Mar 22, 2026
Search Tags:#reinforcement-learning#slime#rl#sglang#post-training#megatron-lm
One-click install
npx skills add https://github.com/nelohenriq/hermes-agent-plus --skill slime-rl-training-nelohenriq
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/nelohenriq/hermes-agent-plus/tree/main/skills/mlops/training/slime
Command: npx skills add https://github.com/nelohenriq/hermes-agent-plus --skill slime-rl-training-nelohenriq

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Post-training reinforcement learning for large language models often involves complex orchestration of data generation, rollout, and policy optimization. This skill provides a guided, slime-based framework to simplify building and operating RL pipelines, reducing integration overhead and accelerating experimentation.

Core Features & Use Cases

  • End-to-end slime RL workflow for GLM, Qwen, Llama families with Megatron-LM integration and SGLang rollout.
  • Flexible data buffering, multi-turn interactions, and custom reward/generate hooks to support RLHF-style refinement.
  • Quick-start examples and references to model scripts to reproduce experiments and scale evaluations on multi-GPU clusters.

Quick Start

Install slime RL tooling, configure your model and data, and start the slime RL training pipeline with a single command to begin post-training reinforcement learning.

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 using Megatron-LM and SGLang?▼

Reinforcement learning post-training for large language models using Megatron-LM and SGLang is orchestrated through the slime framework, which integrates data collection, SGLang-based rollout generation, and policy optimization. You configure your model and data, then execute a single command to start the pipeline.

Can I use the slime framework to run RL post-training on Qwen, GLM, and Llama model families?▼

Yes, the slime framework supports multi-model compatibility for RL post-training across GLM, Qwen, and Llama families. It provides prebuilt scripts and references for model configuration, enabling integrated data collection and policy optimization on multi-GPU clusters.

How does slime handle rollout generation and custom reward hooks during RLHF-style refinement?▼

Slime handles rollout generation during RLHF-style refinement using SGLang-based integration and flexible data buffering. It supports multi-turn interactions and allows you to define custom generate and reward hooks to guide policy optimization.

What dependencies are required to run slime-based reinforcement learning workflows?▼

To run slime-based reinforcement learning workflows, you need Megatron-LM for model integration, SGLang for rollout generation, and the slime runtime environment. These dependencies support configurable rollout parameters and evaluation scaling on multi-GPU clusters.

What is the best way to scale evaluations and reproduce RL post-training experiments on multi-GPU clusters?▼

The best way to scale evaluations and reproduce RL post-training experiments on multi-GPU clusters is by using the slime framework's prebuilt scripts and references. These tools streamline Megatron-LM integration and SGLang-based rollout for accelerated experimentation.

Related Skills

Explore More Education & Research Skills→
ovachieverovachiever
52

grpo-rl-training

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

Community
Advanced
tbroadleytbroadley

read-inspect-eval

Extract samples, messages, events, and metrics from Inspect AI .eval logs.

Community
Intermediate
nahisahonahisaho
69

ai-ml-engineer

Develop end-to-end ML models with training, evaluation, deployment, and MLOps.

Community
Advanced
SignmanalSignmanal
1

slime-rl-training

Coordinate distributed Megatron-LM training and SGLang inference for RL post-training.

Community
Advanced