skills.rest
Signmanal
SignmanalCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

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

1|Updated Jun 25, 2026
Search Tags:#glm#reinforcement-learning#sglang#llm-training#grpo#megatron-lm#rl-post-training
One-click install
npx skills add https://github.com/Signmanal/VIGIL --skill slime-rl-training-signmanal
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/Signmanal/VIGIL/tree/main/optional-skills/mlops/slime
Command: npx skills add https://github.com/Signmanal/VIGIL --skill slime-rl-training-signmanal

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sglang-router>=0.2.3, ray, torch>=2.0.0, transformers>=4.40.0, and includes references (resource) components.

What problem does it solve?

This Skill eliminates the complex, error-prone process of manually coordinating distributed Megatron-LM training, high-throughput SGLang inference, and custom reinforcement learning data pipelines for large language model post-training, which traditionally requires deep expertise in multiple disparate frameworks and extensive configuration tuning.

Core Features & Use Cases

  • Integrated Megatron-LM + SGLang Workflows: Native support for full tensor, pipeline, and sequence parallelism during training, paired with high-throughput rollout generation via SGLang router for RL use cases.
  • Flexible RL Algorithm Support: Built-in configurations for GRPO, PPO, GSPO, and Reinforce++ for post-training GLM, Qwen, DeepSeek, and Llama model families, with custom reward model and multi-turn agentic training capabilities.
  • Use Case: For ML teams training reasoning or tool-use models, this skill removes the need to build and maintain custom glue code between training and inference systems, reducing setup time from weeks to hours.

Quick Start

Use the slime-rl-training skill to launch a GRPO post-training run for your Qwen3-4B model using your local JSONL prompt 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 coordinate distributed Megatron-LM training with SGLang inference for RL post-training?▼

You can coordinate distributed Megatron-LM training and high-throughput SGLang inference by using a framework that natively pairs tensor, pipeline, and sequence parallelism with SGLang router rollout generation for RL post-training.

Can I use GRPO and PPO for post-training GLM and Qwen models?▼

Yes, you can use GRPO, PPO, GSPO, and Reinforce++ algorithms for post-training GLM, Qwen, DeepSeek, and Llama model families, including support for custom reward modeling and multi-turn agentic training.

What is the best way to scale reinforcement learning workflows for large language models?▼

The best way to scale RL workflows is to use integrated distributed training frameworks that support scalable RL scaling, flexible data pipeline management, and off-policy data buffering without requiring custom glue code.

Does slime-rl-training support multi-turn agentic training and custom reward functions?▼

Yes, slime-rl-training supports multi-turn agentic training and custom reward modeling, allowing integration of custom generation functions and flexible data pipeline management for reasoning and tool-use tasks.

What dependencies do I need to run large language model reinforcement learning post-training?▼

You need sglang-router, ray, torch, and transformers to run large language model reinforcement learning post-training, ensuring your environment supports coordinated distributed training and high-throughput inference.

Why does manually setting up LLM RL post-training pipelines fail?▼

Manually setting up LLM RL post-training pipelines fails because it requires deep expertise in disparate frameworks to coordinate distributed Megatron-LM training, SGLang inference, and custom data pipelines without integrated glue code.

Related Skills

Explore More Data & Analytics Skills→
richardnguyen0715richardnguyen0715

slime-rl-training

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

Community
Advanced
GarrettRoiGarrettRoi

slime-rl-training

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

Community
Advanced
projectedanxprojectedanx
1

slime-rl-training

Train reasoning models with Megatron-LM and SGLang for RL post-training.

Community
Advanced
Ced3-hanCed3-han

openrlhf-training

Run distributed RLHF training for large language models with Ray and vLLM.

Community
Advanced