Language Plugin for NLLB200 language translator.

The NLLB200Translator is a translation service that uses the NLLB-200 models to translate text between different languages. It integrates with the OVOS framework and utilizes CTranslate2 for fast and efficient translation.

Below are the available model options for the NLLB-200 Translator:

The plugin can be used within the OVOS framework to translate utterances or texts on demand. Below is an example configuration for integrating the translator within OVOS.

Add the following configuration to one of the configuration files (e.g., ~/.config/mycroft/mycroft.conf):

{
  "language": {
    "translation_module": "ovos-translate-plugin-nllb",
    "ovos-translate-plugin-nllb": {
      "model": "nllb-200_600M_int8"
    }
  }
}

Here's an example of how to use the NLLB200Translator in a Python script:

from ovos_translate_plugin_nllb import NLLB200Translator

src = "es"
tgt = "en-us"
tx = NLLB200Translator(config={"model": "nllb-200_600M_int8"})

utts = ["Hola Mundo"]
print("Translations:", tx.translate(utts, tgt, src))

utts = "hello world"
print("Translations:", tx.translate(utts, src, tgt))

To use models hosted on the HuggingFace Hub, configure the model parameter with the respective HuggingFace model ID. The NLLB200Translator class handles downloading and loading the model from HuggingFace.

• beam_size: Configure the beam size used for translation to balance between translation quality and speed.
• device: Specify the device type, either 'cpu' or 'cuda' (for GPU).

Example:

from ovos_translate_plugin_nllb import NLLB200Translator
tx = NLLB200Translator(config={
    "model": "nllb-200_600M_int8",
    "beam_size": 5,
    "device": "cuda"
})

To leverage GPU acceleration with CUDA, configure the NLLB200Translator to use the cuda device. This can significantly speed up the translation process, especially for large batches or longer texts.

Ensure you have a CUDA-compatible GPU and the necessary CUDA drivers installed on your system. Refer to the CUDA Installation Guide for more details.

Here's a complete example demonstrating how to configure and use the NLLB200Translator with CUDA:

from ovos_translate_plugin_nllb import NLLB200Translator

if __name__ == "__main__":
    src = "es"
    tgt = "en-us"
    tx = NLLB200Translator(config={
        "model": "nllb-200-3.3B-int8",
        "beam_size": 5,
        "device": "cuda"
    })
    
    utts = ["Hola Mundo"]
    print("Translations:", tx.translate(utts, tgt, src))
    
    utts = "hello world"
    print("Translations:", tx.translate(utts, src, tgt))