sledgx/sentence-encoder repository overview

⁠sentence-encoder

Universal Sentence Encoder Multilingual as a service.

⁠About the service

The service allows you to transform a variable length text sentence into 512 dimensional vector. This is useful for performing ML experiments in the embeddings⁠ space. The service also allows you to calculate the similarity of two texts with a score ranging from 0 to 1.

⁠How to use

All of the following examples will bring you the service on http://localhost:8080.

⁠Basic usage

This will run the service with the default option:

docker run -d \
    --name sentence-encoder \
    -p 8080:80 \
    sledgx/sentence-encoder

⁠Usage with custom log

You can define the logs verbosity in the LOG_LEVEL environment variable:

docker run -d \
    --name sentence-encoder \
    -e LOG_LEVEL=debug \
    -p 8080:80 \
    sledgx/sentence-encoder

Accepted values are error, warning, info, debug and notset, default is info.

⁠How to make service requests

The service exposes two endpoints with different functionalities.

⁠Encode

With this method you can convert a text into an embedding vector. You can POST the text in json or form data format:

curl -X POST http://localhost:8080/encode \
    -H 'Content-Type: application/json' \
    -d '{"text":"Hello, World!"}'

or

curl -X POST http://localhost:8080/encode \
    -H 'Content-Type: multipart/form-data'
    -F 'text=Hello, World!'

⁠Similarity

With this method you can obtain a similarity score between two texts. You can POST both texts in json format or form data:

curl -X POST http://localhost:8080/similarity \
    -H 'Content-Type: application/json' \
    -d '{"left_text":"Hello everybody","right_text":"Ciao a tutti"}'

or

curl -X POST http://localhost:8080/similarity \
    -H 'Content-Type: multipart/form-data' \
    -F 'left_text=Hello everybody' \
    -F 'right_text=Ciao a tutti'

⁠License

Released under the MIT License⁠.

Universal Sentence Encoder Multilingual model is owned by Google, please refer to this link⁠ to get all licensing information.

As with all Docker images, these likely also contain other software which may be under other licenses (such as Bash, etc from the base distribution, along with any direct or indirect dependencies of the primary software being contained).