Fine-Tuned MarianMT: English ↔ Hindi

Model Description

This is a fine-tuned version of Helsinki-NLP/opus-mt-en-hi trained on the IIT Bombay English–Hindi Parallel Corpus.
It demonstrates English ↔ Hindi translation with improved performance on a small subset for demonstration purposes.

• Base Model: Helsinki-NLP/opus-mt-en-hi
• Languages: English, Hindi
• Pipeline Tag: translation
• Library: 🤗 Transformers

---

Use Case

• Academic bilingual communication
• General English ↔ Hindi translation
• Demonstration of fine-tuning workflow for machine translation

---

Dataset

• Name: IIT Bombay English–Hindi Parallel Corpus (cfilt/iitb-english-hindi)
• Subset Used: 500 training samples, 200 validation samples
• Reason: Clean, aligned bilingual pairs widely used in MT research

---

Training Setup

• Frameworks: PyTorch + Hugging Face Transformers + Datasets
• Epochs: 1
• Batch Size: 8
• Max Length: 64 tokens
• Execution: CPU-only

---

Evaluation

• Metrics: BLEU score, validation loss
• Results:
  • BLEU score: 6.09
  • Validation loss: 4.4919

---

How to Use

from transformers import MarianMTModel, MarianTokenizer

model = MarianMTModel.from_pretrained("sag-9755/Fine_Tune_MarianMT_English_to_Hindi")
tokenizer = MarianTokenizer.from_pretrained("sag-9755/Fine_Tune_MarianMT_English_to_Hindi")

text = "Hello, how are you?"
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=64)
outputs = model.generate(**inputs, num_beams=4, max_length=64)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))