title	Localized_lang_Translator
app_file	app.py
sdk	gradio
sdk_version	5.49.1

NLLB Language Translator (Apple Silicon Optimized)

This project is a fork of Rohan Bagulwar’s Quantized Language Translator, originally designed for quantized CPU usage.

This modified version has been updated to run efficiently on Apple Silicon Macs (M1/M2/M4) using MPS (Metal Performance Shaders) for GPU acceleration, and no longer relies on CUDA or quantization.

---

🔧 Key Changes

• ✅ Removed bitsandbytes and 8-bit quantization (not compatible with macOS without CoreML).
• ✅ Switched to Apple's MPS backend for GPU acceleration on Mac.
• ✅ Compatible with MacBook Air/Pro (M1/M2/M3/M4).
• ✅ Updated app.py and requirements.txt to reflect Apple Silicon optimizations.
• ✅ Note on the Model: This app uses Meta’s official facebook/nllb-200-distilled-600M model, running in full precision (float32) on Apple Silicon via PyTorch’s MPS backend.

---

🚀 Features

• Apple Silicon Optimized: Runs on Mac GPU using PyTorch MPS backend.
• Multi-language Support: Translate between many language pairs using NLLB.
• Fast Inference on Mac: Leveraging the GPU gives significant speedup over CPU.
• Gradio Interface: Easy-to-use web UI for testing and demo.

---

✅ Setup Instructions

1. Clone the repository

git clone https://github.com/sibi-seeni/nllb_localized_lang_translator.git
cd nllb_localized_lang_translator

2. Create and activate a virtual environment

python3 -m venv myenv
source myenv/bin/activate

3. Install dependencies

pip install -r requirements.txt

Make sure you're using Python ≥3.8 and a recent version of pip.

4. Run the Gradio app:

python app.py

🌍 Supported Languages

Languages include (but are not limited to):

• Tamil 🇮🇳
• Hindi 🇮🇳
• French 🇫🇷
• Spanish 🇪🇸
• German 🇩🇪
• Arabic 🇸🇦

A complete list of supported languages is available in the web app's dropdown menus.

References

• Original Hugging Face Space by Rohan
• NLLB: No Language Left Behind (Meta AI)
• Hugging Face Transformers Documentation
• PyTorch MPS Documentation

Acknowledgements

Thanks to the Hugging Face and Meta AI teams for their amazing models and tooling.