m1guelpf / tinyvector Goto Github PK

tinyvector - a tiny embedding database in pure Rust

• Tiny: It's in the name. It's literally just an axum server. Extremely easy to customize, around 600 lines of code.
• Fast: Tinyvector should have comparable speed to advanced vector databases when it comes on small to medium datasets, and slightly better accuracy.
• Vertically Scales: Tinyvector stores all indexes in memory for fast querying. Very easy to scale up to 100 million+ vector dimensions without issue.
• Open Source: MIT Licensed, free forever.

• Powerful Queries: Allow filtering by the provided vector metadata without slowing the search down.
• Integrated Models: Soon you won't have to bring your own vectors, just generate them on the server automaticaly. Aiming to support support SBert, Hugging Face models, OpenAI, Cohere, etc.
• Typescript/Python Libraries: Should be able to auto-generate pretty good clients using the included OpenAPI schema.

We provide a lightweight Docker container that you can run anywhere. It only takes one command to get up and running with the latest changes:

docker run \
  -p 8000:8000 \
  ghcr.io/m1guelpf/tinyvector:edge

Note When running via Docker Compose or Kubernetes, make sure to bind a volume to /tinyvector/storage for persistence. This is handled automatically in the command above.

You can build tinyvector from the latest tagged release by running cargo install tinyvector (you might need to install Rust first). Then, run tinyvector to start up the server.

You can also build it from the latest commit by cloning the repo and running cargo build --release, and run it with ./target/release/tinyvector.

Most vector databases are overkill for simple setups. For example:

• Using embeddings to chat with your documents. Most document search is nowhere close to what you'd need to justify accelerating search speed with HNSW or FAISS.
• Doing search for your website or store. Unless you're selling 1,000,000 items, you don't need Pinecone.

Embeddings are a way to compare similar things, in the same way humans compare similar things, by converting text into a small list of numbers. Similar pieces of text will have similar numbers, different ones have very different numbers.

Read OpenAI's explanation.

• Will Depue's tinyvector (python+sqlite+numpy) inspired me to build a vector database from scratch (and borrow the name). Will also contributed plenty of ideas to optimize performance.

This project is open-sourced under the MIT license. See the License file for more information.