🌍 NLLB-200 Kabardian Language Translator
This app uses the panagoa/nllb-200-distilled-600M-kbd-v0.1 model, which is fine-tuned for Kabardian language translation. It supports translation between Kabardian and several major languages including English, Russian, Turkish, German, French, Spanish, and Portuguese.
The model is a distilled version (600M parameters) of the larger NLLB models, optimized for efficiency while maintaining good translation quality.
Source Language
Target Language
10 200
Translation Generation Parameters
These parameters control how the model generates translations. Adjust them to fine-tune the translation quality and style:
- Beam Size: Controls how many different translation candidates the model considers. Higher values generally produce better quality but are slower.
- Temperature: Controls randomness in generation. Higher values (>1.0) produce more diverse and creative translations, while lower values produce more deterministic ones.
- Top-p: Controls diversity via nucleus sampling - only the most likely tokens are considered. Lower values make the output more focused.
- Length Penalty: Controls the preference for longer or shorter translations. Values >1.0 favor longer outputs.
- Repetition Penalty: Reduces the likelihood of repeating the same words. Higher values more strongly discourage repetition.
1 10
0.1 2
0.1 1
0.1 2
1 3
Example Translations
| Source Text | Source Language | Target Language |
|---|
About the Model
This translator uses a distilled version of Meta AI's NLLB-200 model, fine-tuned specifically for Kabardian language.
Key features:
- Based on the 600M parameter distilled NLLB model (smaller and faster than the full 1.3B model)
- Optimized for Kabardian language translation
- Supports major languages: Kabardian, English, Russian, Turkish, German, French, Spanish, and Portuguese
- Suitable for real-time translation with lower latency requirements
Limitations:
- This is a distilled model, so it may not match the quality of larger models for complex translations
- Not recommended for critical or professional translation needs
- Performance varies across language pairs and domains