skills.rest
LynxLabVN
LynxLabVNOFFICIAL
Explore all skills by this author →
GitHub

slime-rl-training

Automate post-training reinforcement learning optimization for large language models using the slime framework.

Updated Jul 3, 2026
Search Tags:#LLM#GLM#RL#SGLang#post-training#Megatron-LM#Qwen3
One-click install
npx skills add https://github.com/LynxLabVN/office-agent --skill slime-rl-training-lynxlabvn
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/LynxLabVN/office-agent/tree/main/agent-core/optional-skills/mlops/slime
Command: npx skills add https://github.com/LynxLabVN/office-agent --skill slime-rl-training-lynxlabvn

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve?

This Skill addresses the challenge of efficiently training large language models (LLMs) using reinforcement learning (RL) techniques with the slime framework. It streamlines the process of integrating RL with LLMs and provides flexibility in data generation and workflow design.

Core Features & Use Cases

  • High-Throughput Rollout Generation: Utilizes SGLang for rapid response generation, suitable for large-scale RL tasks.
  • Custom Data Generation: Offers tools for flexible data buffering and generation, ideal for bespoke workflows.
  • Model Support: Supports various LLMs including GLM, Qwen3, DeepSeek V3/R1, and Llama 3.
  • Use Case: For a company training a custom LLM for a specific application, slime can be used to optimize the model for better performance in that domain.

Quick Start

Execute the following command to begin RL training with slime: python train.py --actor-num-nodes 1 --actor-num-gpus-per-node 4 --rollout-num-gpus 4 ... --num-rollout 3000 ... --prompt-data /path/to/data.jsonl

Frequently Asked Questions about slime-rl-training

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

FAQPage Schema
How do I optimize LLM post-training with reinforcement learning using slime?▼

To optimize LLM post-training with reinforcement learning using slime, you automate rollout generation and custom data workflows by executing training scripts with specified GPU allocations and prompt data paths.

Can I train Qwen3 or GLM models using slime for reinforcement learning?▼

Yes, slime supports reinforcement learning training for Qwen3, GLM, DeepSeek V3/R1, and Llama 3 models, enabling custom domain-specific optimization for these large language models.

Do I need Ray and SGLang for high-throughput rollout generation in LLM training?▼

Yes, high-throughput rollout generation in LLM training requires SGLang for rapid response generation and Ray for distributed computing, alongside torch and transformers dependencies.

What is the best way to start RL training with slime for large language models?▼

The best way to start RL training with slime is to execute the train.py script, specifying actor nodes, GPUs per node, rollout GPUs, rollout count, and your prompt data path.

Does slime support custom data generation workflows for LLM reinforcement learning?▼

Yes, slime provides tools for flexible custom data buffering and generation, allowing you to design bespoke data workflows for large-scale LLM reinforcement learning tasks.

Why use SGLang for rollout generation in LLM post-training?▼

SGLang is used for rollout generation in LLM post-training to achieve high-throughput response generation, making it suitable for handling large-scale reinforcement learning workloads efficiently.

Related Skills

Explore More Software Engineering Skills→
ovachieverovachiever
52

ray-train

Orchestrate distributed training across clusters with Ray Train and Ray Tune.

Community
Advanced
vasilyu1983vasilyu1983
73

ai-llm-inference

Optimize LLM inference performance and cost with vLLM and quantization.

Community
Advanced
zangjeicyzangjeicy

slime-rl-training

Train LLMs with GRPO and PPO using Megatron-LM and SGLang.

Community
Advanced
zhouboyu-xrealzhouboyu-xreal

slime-rl-training

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

Community
Advanced