Expand description
§Zebra
A vector database for querying meaningfully similar data. The database can be extended to support documents of many data types; at the moment, text, image, and audio data are supported.
§The name
Zebras (like other equines) make a ‘neigh’ sound. The database performs a nearest-neighbour search to find similar data.
§Installation
Zebra is intended for use as a library. You can add it as a dependency to a Rust project with the following command:
cargo add --git "https://github.com/emmyoh/zebra"
Additionally, a command-line interface (CLI) exists for usage in the terminal.
With the Rust toolchain installed, run the following command:
cargo install --git https://github.com/emmyoh/zebra --features="cli"
You should specify the features relevant to your use case. For example, if you’re interested in using the Zebra CLI on an Apple silicon device, run:
cargo install --git https://github.com/emmyoh/zebra --features="cli,accelerate,metal"
§Features
accelerate- Uses Apple’s Accelerate framework when running on Apple operating systems.cuda- Enables GPU support with Nvidia cards.mkl- Uses Intel oneMKL with Intel CPUs and GPUs.metal- Enables GPU support for Apple silicon machines.sixel- Prints images in Sixel format when using the CLI with compatible terminals.avif- Enables querying & inserting AVIF images.cli- Provides a command-line interface to Zebra.
Modules§
- A module for audio database operations.
- A module for database operations regardless of data type.
- A module for distance metrics.
- A module for image database operations.
- A module for embedding models.
- A module for text database operations.
Constants§
- The candidate list size for the HNSW graph. Higher values result in more accurate search results at the expense of slower retrieval speeds. Can be changed after database creation.