aiproject-2025/english-hindi-odia-translator

English to Hindi/Odia Translator

Fine-tuned IndicTrans2 model for English → Hindi and English → Odia translation.

Model Description

This model is based on ai4bharat/indictrans2-en-indic-1B and fine-tuned on a custom English-Odia dataset.

Supported Languages

• Source: English (en)
• Targets: Hindi (hi), Odia (or)

Usage

from transformers import MarianTokenizer, MarianMTModel

# Load model and tokenizer
model_name = "YOUR_HF_USERNAME/english-hindi-odia-translator"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)

# Translate to Hindi
text = "Hello, how are you?"
inputs = tokenizer(f"<2hin> {text}", return_tensors="pt")
outputs = model.generate(**inputs)
translation = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(translation)

# Translate to Odia
inputs = tokenizer(f"<2ori> {text}", return_tensors="pt")
outputs = model.generate(**inputs)
translation = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(translation)

Training Data

• Dataset: Custom English-Odia parallel corpus
• Size: 354 sentence pairs
• Domain: General conversation and technology

Training Details

• Base Model: ai4bharat/indictrans2-en-indic-1B
• Training Framework: Transformers
• Training Time: ~3 epochs
• Hardware: CPU/GPU

Performance

The model provides improved translation quality for English→Odia compared to the base model, especially for domain-specific content.

Limitations

• Limited training data size
• May struggle with very long sentences
• Performance may vary for specialized domains

License

Apache 2.0

Citation

If you use this model, please cite:

@misc{english-hindi-odia-translator,
  title={English to Hindi/Odia Translator},
  author={Your Name},
  year={2025},
  url={https://huggingface.co/YOUR_HF_USERNAME/english-hindi-odia-translator}
}