Instructor Embedding API

This repository contains a lightweight Sanic API for creating embeddings using the Instructor model. It is provided as a Docker container and based on the hkunlp/instructor-large model. The API can be used for versatile purposes, including in applications such as text classification, similarity, or clustering tasks.

For more information about the Instructor model, visit the official links:

• Model on Hugging Face
• GitHub Repository

Quick Start

Prerequisites

• Docker, version 20.10 or newer

Build and Run the API

1. Clone this repository:

   git clone https://github.com/flexchar/instructor-embedding-api.git
   cd instructor-embedding-api

2. Build a Docker image:

   make build

3. Run the Docker container:

   make run

The API will be available at http://127.0.0.1:8000/.

Consume Pre-built Container from GitHub Packages

You can also use the pre-built container available on GitHub Packages:

docker pull ghcr.io/flexchar/instructor-embedding-api:latest
docker run --rm -p 8000:8000 ghcr.io/flexchar/instructor-embedding-api:latest

Use the API

You can use the API to generate embeddings by sending a POST request to http://127.0.0.1:8000/ with a JSON payload in the format:

{
    "input": [instruction_sentence_pairs]
}

instruction_sentence_pairs is a list of pairs, where each pair contains two strings: an instruction and a sentence.

For example:

{
    "input": [
        [
            "Represent the Fitness title:",
            "What is the easiest training plan for a newbie?"
        ]
    ]
}

A valid response will have the following structure:

{
    "model": "hkunlp/instructor-large",
    "data": [embeddings]
}

embeddings is a list of arrays representing the embeddings for the given instruction-sentence pairs.

License

This project is licensed under the MIT License. See the LICENSE file for details.