README.md

Translation API built on Prediction Guard

Example translation app built on top of Prediction Guard, extendable with DeepL, OpenAI, etc.

To run:

1. Copy config_template.yml to config.yml
2. Fill in or modify config.yml as needed/ appropriate
3. Run python main.py

(alternatively run in Docker)

Adding a customer translation engine

This translation service can integrate with custom translation engines, assuming they fulfill the expected API contract. When adding a custom engine, the following kind of entry is needed in the config.yml:

  custom:
    models:
      model_name:
        url: https://your-custom-url
        api_key: your-custom-api-key
        languages:
          - eng
          - fra
          - deu
          - cmn

The config.yml entry should have:

• url: The endpoint location to call the engine
• api_key: The API key for the engine endpoint, which will be added as x-api-key in the API call headers
• languages: List of supported ISO639-3 language codes

The custom translation engine endpoint should expect a JSON body that looks like:

{
    "text": "The sky is blue",
    "model": "nllb",
    "source_lang": "eng",
    "target_lang": "fra"
}

Where:

• text (required): The text that will be translated
• model (required): The name of the model used (as some endpoints will integrate multiple models)
• target_lang (required): The ISO639-3 (three letter) code specifying the target language
• source_lang (optional): The ISO639-3 (three letter) code specifying the source language

The custom translation engine endpoint should respond with a JSON body including a translation field containing the text translation (string).