Detoxify Simple API

A lightweight FastAPI wrapper around the Detoxify library for toxicity detection in text. This API allows you to quickly detect toxic content in text using pre-trained models.

Overview

This project provides a simple REST API for the Detoxify library, which uses deep learning models to identify toxic content in text. The API can detect various types of toxicity, including:

• Toxicity
• Severe toxicity
• Obscenity
• Threat
• Insult
• Identity-based hate

Getting Started

Prerequisites

• Docker and Docker Compose (recommended)
• Python 3.10+ (for local development)

Running with Docker

Using a command

docker run -d \
  --name detoxify-api \
  -v "$(pwd)/.cache/application:/root/.cache" \
  -v "$(pwd)/.cache/packages:/.cached_packages" \
  -p 8000:8000 \
  ghcr.io/ceraia/detoxify-api:latest

Using Docker Compose (Recommended)

The easiest way to run the API is using Docker:

docker-compose up

This will start the API server on port 8000.

Running Locally

1. Install the required dependencies:

pip install -r requirements.txt

2. Start the API server:

uvicorn toxicity_api:app --host 0.0.0.0 --port 8000

API Usage

Endpoint

POST /predict

Request Format

{
  "text": "The text you want to analyze for toxicity"
}

Response Format

{
  "toxicity": 0.05,
  "severe_toxicity": 0.01,
  "obscene": 0.02,
  "threat": 0.01,
  "insult": 0.03,
  "identity_hate": 0.01
}

The values are probabilities between 0 and 1, where higher values indicate higher likelihood of that category of toxicity.

Example

curl -X POST "http://localhost:8000/predict" \
     -H "Content-Type: application/json" \
     -d '{"text":"This is a test message"}'

Project Structure

• toxicity_api.py - FastAPI application that serves the toxicity detection endpoint
• Dockerfile - Configuration for building the Docker image
• docker-compose.yml - Configuration for running the service
• requirements.txt - Python dependencies

Disclaimer

This project is a simple API wrapper around the Detoxify library. All rights to the underlying models, libraries, and resources belong to their respective owners. This project does not claim ownership over any third-party components used, including but not limited to the Detoxify library and its pre-trained models.

• The Detoxify library is developed and maintained by Unitary AI: https://github.com/unitaryai/detoxify
• FastAPI is developed by Sebastián Ramírez: https://github.com/tiangolo/fastapi

License

Please note that the Detoxify models have their own licensing terms. Refer to the Detoxify repository for details on model licensing.