A compact, production-ready multilingual neural machine translation model supporting 20 languages (190 language pairs). Trained on curated OPUS-100 data with synthetic augmentation, knowledge distillation, and neural quality filtering.
1# Translate a sentence
2uts serve --config config/base.yaml
3curl -X POST http://localhost:8000/translate \
4 -H "Content-Type: application/json" \
5 -d '{"text": "Hello world", "source": "en", "target": "es"}'
1from runtime.encoder.universal_encoder import UniversalEncoder
2from runtime.cloud_decoder import OptimizedUniversalDecoder
3
4encoder = UniversalEncoder.from_pretrained("code-with-zeeshan/Universal-Translation-System")
5decoder = OptimizedUniversalDecoder.from_pretrained("code-with-zeeshan/Universal-Translation-System")
6# See docs/API.md for full inference examples
1from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
2
3model = AutoModelForSeq2SeqLM.from_pretrained("code-with-zeeshan/Universal-Translation-System")
4tokenizer = AutoTokenizer.from_pretrained("code-with-zeeshan/Universal-Translation-System")
The model was trained using the
Universal Translation System pipeline:
