HuangOwen/Awesome-LLM-Compression

What it solves

Large Language Models (LLMs) are often massive, making them slow to run and expensive to train. This project provides a curated collection of research and tools designed to make these models smaller, faster, and more efficient without sacrificing too much performance.

How it works

It serves as a comprehensive, organized repository of academic papers and software tools. The content is categorized by technique—such as quantization, pruning, distillation, and KV cache compression—and further sorted by year. Users can browse these categories to find the latest state-of-the-art methods for accelerating model training and inference.

Who it’s for

This project is for researchers, machine learning engineers, and developers who need to optimize LLMs for deployment on resource-constrained hardware, reduce latency in production, or lower the computational costs of training.

Highlights

• Extensive Research Library: Access to hundreds of papers covering surveys, quantization, pruning, distillation, and efficient prompting.
• Tooling Directory: A categorized list of inference engines, quantization libraries, and model compression toolkits.
• Chronological Organization: Papers are grouped by year with collapsible sections, ensuring the most recent advancements are easy to find.
• Practical Focus: Includes links to code repositories, GitHub pages, and model checkpoints for many of the listed techniques.