LLM Summarization API (FastAPI + BART-Large)

This repository provides a functional microservice for Abstractive Text Summarization. It is developed for consistent performance and uses a BART-Large LLM integrated with FastAPI, Docker, and Gunicorn for deployment.

Setup and Running

The project is designed to be run using Docker-Compose.

1. Prerequisites

• Docker
• Docker Compose
• Sufficient RAM (A minimum of 12 GB RAM is recommended for the Docker Engine/VM to safely load the model.)

2. Clone the Repository

git clone abstractive-summarizer-api
cd abstractive-summarizer-api

3. Build and Start the Service

The following command builds the Docker image, caches the model weights inside the image, and starts the API service.

Note: The initial build process may take some time due to the large size of the model. Grab a ☕!

make run

4. Check Status

Ensure the container is running:

make logs

API Usage

The API is available on port 8000 of your local machine.

A. Liveness Check (Service Running)

Checks if the service is alive. ✅

• Endpoint: GET /health
• Successful Response (reflects the StatusOutput schema):

{
    "status": "ok",
    "model_status": "Loading (Awaiting first request)"
}

B. Readiness Check (Model Status)

Checks the precise loading status of the AI model. This is the official readiness endpoint.

• Endpoint: GET /api/v1/status
• Successful Response: Returns {"model_status": "Loaded"} once the model is in memory.

C. Text Summarization (Inference)

Used to summarize a long text.

• Endpoint: POST /api/v1/summarize
• Content-Type: application/json
• Request Body (Note: Pydantic enforces minimum text length):

{
  "text": "The long text goes here...",
  "min_length": 30,
  "max_length": 150
}

• Successful Response:

{
  "summary": "The generated, concise, and coherent summary text."
}

D. Interactive Documentation (Swagger UI)

You can view all API schemas and the interactive testing interface in your browser: 🔎

http://localhost:8000/docs

Quality Assurance & Testing

1. Code Formatting

Checks and automatically fixes code styling (using Black) and import ordering (using Isort).

make format

2. Run Tests

Executes unit tests using pytest to verify core functionality, DI, and error handling. The AI model is mocked during these tests for speed and reliability.

make test

Stopping the Service

Stops and removes all containers and the network created by Docker Compose:

make clean