fast-stager/STOP

STOP Classifier API

This Hugging Face Space hosts a low-latency text classification service deployed with Docker and FastAPI.

The service uses a highly efficient Linear Support Vector Machine (SVM) model trained on text features extracted via TF-IDF to classify messages as either intending to end communication (STOP) or not (NOT_STOP). As confirmed by the training script, the SVM model provides millisecond-level inference, which is ideal for the required low-latency API.

Project Structure

The deployment uses the following structure:

.
├── app.py           
├── Dockerfile       
├── requirements.txt 
├── README.md        
└── checkpoint/
    ├── tfidf_vectorizer.pkl
    └── svm_stop_classifier.pkl

API Endpoints

The FastAPI application provides two primary endpoints for prediction:

1. Health Check (GET)

• —Path: /
• —Method: GET
• —Description: A simple endpoint to confirm the service is running and the models are loaded.

2. Single Prediction (GET)

• —Path: /predict?text=<your_text>
• —Method: GET
• —Description: Classifies a single text string passed as a query parameter. This is suitable for quick, individual queries.
• —Example Query: /predict?text=please%20discontinue%20all%20contact

3. Batch Prediction (POST)

• —Path: /predict
• —Method: POST
• —Description: Classifies a list of text strings in a single request. This is the recommended approach for high-throughput, low-latency production use cases due to reduced overhead.
• —Request Body (JSON):

    {
      "texts": [
        "do not ever text me again",
        "I will stop by your office tomorrow"
      ]
    }