faiss-api

Simple faiss API for index search with flask and docker

About Faiss

Faiss is a tool created by the Facebook Research team to efficiently get vector similarities, more info on https://github.com/facebookresearch/faiss/

How to use

Export your faiss index to .index file like:

import faiss
index = ... #your index logic
index.train(data)
index.add(data)
faiss.write_index(index, "latest.index")
#optional - save the sequencial list of keys
import numpy as np
np.save('keys.npy', np.array(keys_list))

clone this repository
add the latest.index in the app root path
add the keys.npy in the app root path (optional)

Run local with Docker

Build docker image (in the app root path)

docker build -t samuelhei/faiss-api:latest .

Run Image

docker run -e LD_LIBRARY_PATH='/usr/local/lib' -p 5000:5000 samuelhei/faiss-api:latest

Open a new terminal and get the docker IP

$ docker ps $ docker inspect | grep "IPAddress"

Change the <container id> part with the id returned on the docker ps command, the returned IP address will be your host, using the port 5000 like http://127.0.0.2:5000

Deploy on AWS beanstalk

Install the AWS client (if not installed)

$ pip install awscli $ pip install awsebcli

Set your preferences of the region and set Docker as Environment

$ eb init

Create and Deploy application (This will use t2.micro as default)

$ eb create -d --envvars LD_LIBRARY_PATH=/usr/local/lib

When you update your application then you can redeploy

$ eb deploy

Test the endpoint

Open in your browser <host>/ping, if everything is ok it will be returned the following result: {"ping":"pong"}

Getting search results from the index

Send a POST method to <host>/get_similarities with the following parms:
- vector as text like 0.10359007865190506,0.25027135014533997,0.6511507034301758,0.501607358455658,0.39755818247795105,0.007388297468423843,0.33636972308158875,0.1718718260526657,0.1494559347629547,0.23654937744140625,0.39706116914749146
- n as the number of itens to return (default value 5)