Build a Faiss service instantly. Faiss-instant will simply load existing Faiss index (and the corresponding ID mapping) and provide the search service via POST request.

New features: Now Faiss-instant also provides the toolkit for encoding texts into embeddings via SBERT models and indexing the embeddings into a Faiss ANN index. One just needs to install the toolkit via

pip install faiss-instant

and try this example.

First, one needs to put the resource files (the ID mapping and the Faiss index, please refer to resources/README.md) under the folder ./resources:

make download  # This will download example resource files. The example index comes from building a SQ index (QT_8bit_uniform) on a 10K-document version of the NQ corpus (dpr-single-nq-base was used for encoding). For other indices, please find under https://public.ukp.informatik.tu-darmstadt.de/kwang/faiss-instant/.

Then, one needs to start the faiss-instant service via docker:

docker pull kwang2049/faiss-instant  # Or `make pull`; or `make build` to build the docker image
docker ru