Documentation

Twinkle Documentation

Welcome to the Twinkle documentation! Twinkle is a lightweight, client-server training framework engineered with modular, high-cohesion interfaces for LLM training.

Whether you are a researcher customizing training algorithms, a developer building production training pipelines, or an enterprise deploying Training-as-a-Service, this documentation will help you get started.

Quick Start

Install and run your first training in 5 minutes

Concepts & Architecture

Deep dive into architecture, runtime modes, and deployment

Tutorials

Step-by-step guides: Training Guide, Qwen3.5, Embedding Training

Components

Dataset, Model, Sampler, Reward and more

API Reference

Component APIs and customization