跳转至

LLM Project Documentation

Welcome to the LLM project documentation. This index provides quick navigation to all available guides and references.

Document Description
Usage Guide Training, inference, and serving instructions
Architecture System design and component overview
Tutorial Step-by-step CPU LLM training tutorial
Development Setting up the development environment
FAQ Frequently asked questions
Troubleshooting Common issues and solutions

Guides

In-depth guides for specific features:

Guide Description
Fine-Tuning (LoRA/QLoRA) Parameter-efficient fine-tuning methods
Inference Optimization KVCache, GQA, sliding window attention

Training Framework

Detailed documentation for the training system:

Document Description
Overview Training framework introduction
Components Core training components
Training Flow End-to-end training process
Callbacks Callback system for extensibility
Configuration Configuration guide
Extending How to extend the framework
DDP Deep Dive Distributed training details
Troubleshooting Training-specific issues

Architecture Decision Records (ADR)

Design decisions and their rationale:

ADR Topic
001 Grouped Query Attention (GQA)
002 SwiGLU Activation
003 Using prek for Git hooks
004 Using ty for type checking

Deep Dives

In-depth technical explorations:


Contributing: See CONTRIBUTING.md for contribution guidelines.

Issues: Report bugs at GitHub Issues.