|
|
--- |
|
|
title: NLLB200 |
|
|
emoji: ๐ |
|
|
colorFrom: blue |
|
|
colorTo: purple |
|
|
sdk: gradio |
|
|
sdk_version: 4.44.0 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
license: apache-2.0 |
|
|
--- |
|
|
|
|
|
# NLLB-200 Translation API |
|
|
|
|
|
High-quality translation API powered by Meta's **No Language Left Behind (NLLB-200)** model. |
|
|
|
|
|
## Features |
|
|
|
|
|
- โ
**200 Languages Supported** - Direct translation between any language pair |
|
|
- โ
**44% Better Quality** - Compared to previous translation models |
|
|
- โ
**+70% for Complex Languages** - Especially Arabic, Hindi, and other low-resource languages |
|
|
- โ
**No Pivot Translation** - Direct translation without going through English |
|
|
- โ
**Cached Results** - Faster repeated translations |
|
|
- โ
**API Access** - Use via Gradio Client or HTTP |
|
|
|
|
|
## Supported Languages (Sample) |
|
|
|
|
|
Arabic, English, French, Spanish, German, Italian, Portuguese, Russian, Japanese, Korean, Chinese, Hindi, Turkish, Dutch, Polish, Swedish, Indonesian, Vietnamese, Thai, Ukrainian, Romanian, Greek, Hebrew, and more! |
|
|
|
|
|
## Usage |
|
|
|
|
|
### Web Interface |
|
|
|
|
|
Visit the Space and use the interactive interface to translate text between any supported languages. |
|
|
|
|
|
### API |
|
|
|
|
|
```python |
|
|
from gradio_client import Client |
|
|
|
|
|
client = Client("TGPro1/NLLB200") |
|
|
result = client.predict( |
|
|
"Hello, world!", # text |
|
|
"English", # source language |
|
|
"Arabic", # target language |
|
|
api_name="/predict" |
|
|
) |
|
|
print(result) # ู
ุฑุญุจุง ุจุงูุนุงูู
! |
|
|
``` |
|
|
|
|
|
## Model |
|
|
|
|
|
- **Model**: `facebook/nllb-200-distilled-600M` |
|
|
- **Parameters**: 600M |
|
|
- **Size**: ~2.4GB |
|
|
- **Languages**: 200 |
|
|
|
|
|
## Performance |
|
|
|
|
|
- **Quality**: State-of-the-art for low-resource languages |
|
|
- **Speed**: Fast inference with distilled model |
|
|
- **Cache**: LRU cache for frequently translated phrases |
|
|
|
|
|
## Credits |
|
|
|
|
|
- **Model**: Meta AI Research - [NLLB Project](https://ai.meta.com/research/no-language-left-behind/) |
|
|
- **Benchmark**: FLORES-200 |
|
|
- **License**: Apache 2.0 |
|
|
|
|
|
## Links |
|
|
|
|
|
- [Model Card](https://huggingface.co/facebook/nllb-200-distilled-600M) |
|
|
- [Research Paper](https://arxiv.org/abs/2207.04672) |
|
|
- [Full Language List](https://github.com/facebookresearch/flores/blob/main/flores200/README.md) |
|
|
|
