Re-Ranker API

Overview

The Re-Ranker API is a simple API for ranking documents based on their relevance to a given question. As a note of importance, this project assumes a secure environment where sensitive data will not be handled.

No additional security measures are included in this API, as it is designed to operate within a trusted perimeter.

This project uses FastAPI and Sentence Transformers to rank documents using a cross-encoder model.

How to use

Prerequisites

• Python 3.11 or greater
• pip
• virtual environment (recommended)
• A secure environment

To get started with the Re-Ranker API, you'll need to have Python 3.11 or greater installed on your system. Once you have Python set up, follow these steps:

Step 1: Create and activate a Virtual Environment

The best way to manage dependencies for this project is by using a virtual environment (venv). To create a new venv, run the following command in your terminal:

python -m venv myenv

Replace myenv with the name you want to give your virtual environment. Then, activate it:

source myenv/bin/activate

Step 2: Install dependencies

With your venv activated, install the required packages by running:

pip install -r requirements.txt

This will install all the necessary dependencies for the project.

Step 3: Start the API

Once you have the dependencies installed, start the API by running:

python3 -m uvicorn main:app --host localhost --port 8000

This will start the API on port 8000 and make it available at http://localhost:8000.

Step 4: Send a POST request to rank documents

To send a POST request to the /rank endpoint, you can use any HTTP client tool such as curl or Postman. Here's an example of how you could do this using curl:

curl -X POST \
  http://localhost:8000/rank \
   -H 'Content-Type: application/json' \
   -d '{"question": "What is the capital of France?", "documents": ["The Eiffel Tower is located in Paris,
France.", "Montreal is the capital of Canada."]}''

When you send a POST request to the /rank endpoint, you can expect to receive a JSON object in the following format sorted on the scores, higher is better:

[
  {"corpus_id": 0, "score": "0.843489"},
  {"corpus_id": 2, "score": "0.725346"},
  {"corpus_id": 1, "score": "0.435645"}
]

This JSON object contains an array of objects, each representing a document with its corresponding relevance score. The corpus_id field is the ID of the document sent in the request, and the score field is the calculated relevance score between 0 and 1.