CUHNSWPLUS

It is an improved fork of cuhnsw, featuring enhanced performance and new functionalities.

Improvements

• Multithreading: Parallel loading of graph vectors for faster processing.
• Performance: Reuses allocated device_vector to reduce overhead.
• Disk Caching: Efficient handling of large datasets using disk caching.
• API Support: Added support for AddPoint and AddPoints functions.

Cloning

  git clone --recursive https://github.com/n0tank3sh/cuhnswplus.git

Installation

Python:

pip install .

C++:

cmake -S . -B build
cd build
make install -j`nproc`

Usage:

Copy the example config from examples/examples_config.json

Python

Refer to the examples folder for detailed usage examples.

CPP

Refer to the tests folder for detailed usage examples.

Contributing

We welcome contributions! Please feel free to open issues, submit pull requests, or provide feedback.