skills.rest
twjarviszyra-web
twjarviszyra-webCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Orchestrate reinforcement learning post-training workflows for large language models.

3|Updated Apr 4, 2026
Search Tags:#reinforcement learning#sglang#grpo#post-training#megatron-lm#llm training#agentic training
One-click install
npx skills add https://github.com/twjarviszyra-web/hermes-unbound --skill slime-rl-training-twjarviszyra-web
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/twjarviszyra-web/hermes-unbound/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/twjarviszyra-web/hermes-unbound --skill slime-rl-training-twjarviszyra-web

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill helps teams run large-language-model post-training without stitching together a custom reinforcement learning stack, making it easier to train, evaluate, and iterate on advanced model behavior at scale.

Core Features & Use Cases

  • Megatron-LM training: Coordinates distributed actor, critic, and checkpoint workflows for high-throughput RL fine-tuning.
  • SGLang rollout generation: Uses router-backed inference for fast sample generation, multi-turn interactions, and reward collection.
  • Model and workflow support: Fits GLM, Qwen, DeepSeek, Llama, and other supported families for GRPO, PPO, GSPO, and custom agentic training pipelines.
  • Use case: A research team can launch GRPO training on a reasoning model, plug in a custom reward function, and monitor rollout quality with built-in evaluation and troubleshooting guidance.

Quick Start

Ask the assistant to set up a slime RL training workflow for your model, dataset, and hardware constraints.

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 reinforcement learning post-training for large language models?▼

Reinforcement learning post-training for large language models is orchestrated by coordinating distributed actor and critic workflows with Megatron-LM and using SGLang for fast rollout generation. This setup manages GRPO, PPO, GSPO, and multi-turn agentic training scenarios.

Does SGLang work with Megatron-LM for RL rollout generation?▼

Yes, SGLang works with Megatron-LM by providing router-backed inference for fast sample generation, multi-turn interactions, and reward collection, while Megatron-LM handles the distributed training and checkpoint synchronization.

Can I use GRPO training on a Qwen or DeepSeek reasoning model?▼

Yes, you can use GRPO training on Qwen, DeepSeek, GLM, and Llama model families. The workflow supports plugging in custom reward functions and monitoring rollout quality with built-in evaluation and troubleshooting guidance.

What do I need to set up multi-turn agentic training pipelines?▼

To set up multi-turn agentic training pipelines, you need structured prompt and label data, model configuration scripts, custom generation or reward functions, and distributed training coordination for evaluation, buffering, and checkpoint synchronization.

What's the best way to scale distributed RL fine-tuning without a custom stack?▼

The best way to scale distributed RL fine-tuning without a custom stack is using an orchestration layer that coordinates high-throughput actor and critic workflows across Megatron-LM, handling checkpoint synchronization and rollout buffering automatically.

Why does PPO post-training require custom reward functions?▼

PPO post-training requires custom reward functions to define the target advanced model behavior during rollout generation. The workflow uses these functions with SGLang to collect rewards and guide the RL fine-tuning loop.

Related Skills

Explore More Software Engineering Skills→
JKhyroJKhyro

slime-rl-training

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

Community
Advanced
Jonah-1Jonah-1

eval

Run inference on neural network checkpoints and generate PSNR, SSIM, and LPIPS metrics.

Community
Intermediate
hhhi21ghhhi21g

slime-rl-training

Scale RL post-training for LLMs with Megatron-LM and SGLang.

Community
Advanced
jason660519jason660519

slime-rl-training

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

Community
Advanced