This repository serves as a comprehensive, continuously updated knowledge base for Artificial General Intelligence (AGI), focusing on Large Language Models (LLMs) and AI Generated Content (AIGC). It targets researchers, practitioners, and enthusiasts in the AI field, offering a structured overview of LLM benchmarks, architectures, pipelines, training, fine-tuning, deployment, and applications, alongside a curated list of AIGC tools and concepts.

How It Works

The project organizes information into distinct categories, providing detailed breakdowns of LLM concepts like prompt engineering, RLHF, and long-context handling. It extensively covers Parameter-Efficient Fine-Tuning (PEFT) techniques, including LoRA and its variants, and offers insights into LLM deployment and distributed training frameworks. A significant portion is dedicated to Retrieval-Augmented Generation (RAG), detailing practical implementations with frameworks like LangChain and LlamaIndex, and exploring various vector databases.

Quick Start & Requirements

This is a curated knowledge base, not a runnable software project. Access to the information is via the README.

Highlighted Details

• Extensive benchmark scores for various LLMs across multiple evaluation metrics.
• Detailed categorization of LLMs by architecture, domain, and API availability.
• In-depth explanations and comparisons of numerous PEFT methods, including LoRA, QLoRA, and DoRA.
• Comprehensive overview of RAG implementation, vector databases, and agent frameworks like AutoGen, AutoGPT, and MetaGPT.

Maintenance & Community

The project is actively maintained and encourages community contributions through pull requests and stars. A WeChat group is available for technical exchange.

Licensing & Compatibility

The repository itself does not specify a license. Content is presented for learning and reference, with a disclaimer that some materials are compiled from the internet.

Limitations & Caveats

As a knowledge repository, it does not provide executable code or direct tooling. The information is a compilation and may require further investigation for practical application. Some content is noted as being sourced from the internet, implying potential copyright considerations for redistribution.