app.py

import gradio as gr
from transformers import pipeline
import torch
import logging
import spaces
from typing import Literal, Tuple

# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Automatically detect the available device (CUDA, MPS, or CPU)
if torch.cuda.is_available():
    device = "cuda"
    logger.info("Using CUDA for inference.")
elif torch.backends.mps.is_available():
    device = "mps"
    logger.info("Using MPS for inference.")
else:
    device = "cpu"
    logger.info("Using CPU for inference.")

# Load the translation pipeline with the specified model and detected device
model_checkpoint = "oza75/bm-nllb-1.3B"
translator = pipeline("translation", model=model_checkpoint, device=device, max_length=512)
logger.info("Translation pipeline initialized successfully.")

# Define the languages supported
SOURCE_LANG_OPTIONS = {
    "French": "fra_Latn",
    "English": "eng_Latn",
    "Bambara": "bam_Latn",
    "Bambara With Error": "bam_Error"
}

TARGET_LANG_OPTIONS = {
    "French": "fra_Latn",
    "English": "eng_Latn",
    "Bambara": "bam_Latn"
}


# Define the translation function with typing
@spaces.GPU()
def translate_text(text: str, source_lang: str, target_lang: str) -> str:
    """
    Translate the input text from the source language to the target language using the NLLB model.

    Args:
        text (str): The text to be translated.
        source_lang (str): The source language code (e.g., "fra_Latn", "bam_Error").
        target_lang (str): The target language code (e.g., "eng_Latn", "bam_Latn").

    Returns:
        str: The translated text.
    """
    source_lang, target_lang = SOURCE_LANG_OPTIONS[source_lang], TARGET_LANG_OPTIONS[target_lang]
    logger.info(f"Translating text from {source_lang} to {target_lang}.")
    try:
        # Perform translation using the Hugging Face pipeline
        result = translator(text, src_lang=source_lang, tgt_lang=target_lang)
        translated_text = result[0]['translation_text']
        logger.info("Translation successful.")
        return translated_text
    except Exception as e:
        logger.error(f"Translation failed: {e}")
        return "An error occurred during translation."


# Define the Gradio interface
def build_interface():
    """
    Builds the Gradio interface for translating text between supported languages.

    Returns:
        gr.Interface: The Gradio interface object.
    """
    # Define Gradio input and output components
    text_input = gr.Textbox(lines=5, label="Text to Translate", placeholder="Enter text here...")
    source_lang_input = gr.Dropdown(choices=list(SOURCE_LANG_OPTIONS.keys()), value="French", label="Source Language")
    target_lang_input = gr.Dropdown(choices=list(TARGET_LANG_OPTIONS.keys()), value="Bambara", label="Target Language")
    output_text = gr.Textbox(label="Translated Text")

    # Define the Gradio interface with the translation function
    return gr.Interface(
        fn=translate_text,
        inputs=[text_input, source_lang_input, target_lang_input],
        outputs=output_text,
        title="Bambara NLLB Translation",
        description=(
            "This application uses the NLLB model to translate text between French, English, and Bambara. "
            "The source and target languages should be chosen from the dropdown options. If you encounter "
            "any issues, please check your inputs."
        ),
        examples=[
            ["Thomas Sankara, né le 21 décembre 1949 à Yako (Haute-Volta) et mort assassiné le 15 octobre 1987 à Ouagadougou (Burkina Faso), est un homme d'État voltaïque, chef de l’État de la république de 'Haute-Volta', rebaptisée Burkina Faso, de 1983 à 1987.", "French", "Bambara"],
            ["Good morning", "English", "Bambara"],
            ["- Ɔridinatɛri ye minɛn ye min bɛ se ka porogaramu - A bɛ se ka kunnafoniw mara - A bɛ se ka kunnafoniw sɔrɔ - A bɛ se ka kunnafoniw baara", "Bambara", "French"],
        ]
    )


# Run the Gradio application
if __name__ == "__main__":
    logger.info("Starting the Gradio interface for the Bambara NLLB model.")
    interface = build_interface()
    interface.launch()
    logger.info("Gradio interface running.")