Flask App with Sentence Transformers

This repository contains a Flask application that uses the SentenceTransformer model to compute the similarity between two input sentences. The application is containerized using Docker and is configured to run in both development and production environments using gunicorn.

A Simple Server to Compute the score of similarity between two strings.

Features

• Flask Backend: A simple API built with Flask that accepts two sentences and returns their cosine similarity score.
• Sentence Transformers: Uses the SentenceTransformer model to encode sentences and compute their similarity.
• YAML Configurations: Reads settings such as model name, device configuration, and server options from a YAML file (config.yaml).
• Gunicorn: In production, the app is served using gunicorn for improved performance.

Prerequisites

Docker

• Python 3.x (if you want to run it locally without Docker)
• docker-compose (optional, if you want to use it to manage services)

Getting Started

1. Clone the Repository
2. Modify the config.yaml
3. Build and Run the Docker Container Build the Docker image and run the container:

docker build -t mlserver .
docker run -p 5000:5000 mlserver

By default, the application runs in production mode with gunicorn. 4. Running in Development Mode If you want to run the app in development mode with Flask’s built-in server and auto-reload, override the FLASK_ENV environment variable:

docker run -p 5000:5000 -e FLASK_ENV=development mlserver

5. Accessing the API Once the Docker container is running, you can send a POST request to the server to compute the sentence similarity.

Example curl command:

curl -v -H "Content-type: application/json" --data '{"s1":"This is a Surface Studio Laptop","s2":"That is a car"}' http://127.0.0.1:5000

The response will contain the cosine similarity score between the two sentences.

6. Show the Logs

`docker logs mlserver` or `docker logs -f mlserver`

Development (Without Docker)

To run the application locally without Docker:

1. Install Dependencies First, ensure you have Python 3.x installed, then install the necessary Python packages:

pip install -r requirements.txt

2. Start the Flask Application Start the app using the following command:

FLASK_ENV=development flask run

The server will be available at http://127.0.0.1:5000.

Docker Compose (Optional)

To use docker-compose to manage services, you can create a docker-compose.yml file and run the application with:

docker-compose up --build -d

This will build and start the Flask application along with any additional services you define.

To view the logs using docker-compose, run:

docker-compose logs -f

To restart the docker-compose container, run:

docker-compose down  # Stop the containers
docker-compose up -d  # Start the containers in detached mode

Or simply:

docker-compose restart <service_name>

Configuration

The application reads the following configurations from config.yaml:

• Model Settings: Specify the model name (e.g., paraphrase-MiniLM-L6-v2), device (auto, cuda, or cpu), and precision (float32 or float16).
• Caching: Enable or disable caching for sentence embeddings.
• Server Settings: Configure the host, port, and number of gunicorn workers.

Production

In production, the app is served using gunicorn. You can customize the number of workers by modifying the config.yaml file.

Exposed Ports

The application exposes port 5000 by default. This can be modified in the docker run command or the config.yaml file.

License

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

You can also run ./build-and-run.sh to build the Docker image and run the server.

Support

If you find this useful, consider buy me a cup of coffee, thanks! https://www.buymeacoffee.com/y0BtG5R