Skip to content

Model API

The project provides a REST API built with FastAPI, exposing a /predict endpoint for classifying code comments by their programming language. This API allows developers to send text snippets and receive predicted labels along with model metadata.

Test the API Locally

Tip

Make sure the git repository has been cloned and all dependencies are installed before running the server.

1. Start the Server

Run the FastAPI server locally using Uvicorn:

uvicorn turing.api.app:app --reload
  • The --reload flag enables automatic reloading whenever code changes are detected
  • By default, the server runs at http://127.0.0.1:8000

2. Request Payload

Send a JSON request containing:

  • texts: a list of code comments
  • language: the language context for prediction
{
  "texts": [
    "public void main",
    "def init self"
  ],
  "language": "pharo"
}

3. Response Body

The API returns a JSON response with:

  • labels: predicted labels for each input text
  • predictions: numeric labels for each input text
  • model_info: metadata about the model used for prediction

Example Response

{
  "predictions": [
    [1, 0, 0, 0, 0, 0, 0],
    [1, 0, 0, 0, 0, 0, 0]
  ],
  "labels": [
    ["summary"],
    ["summary"]
  ],
  "model_info": {
    "artifact": "RandomForestTfIdf_java",
    "language": "java"
  }
}

Supported Languages

  • Java: Fine-tuned CodeBERTa model
  • Python: GraphCodeBERT model
  • Pharo: GraphCodeBERT model

Interactive API Documentation

FastAPI automatically provides interactive API documentation:

You can test the API directly from these interfaces by clicking "Try it out" on any endpoint.