54gO/marianmt-en-hi-translator

Model Card for Model ID

MarianMT English → Hindi Translator (Fine-tuned with QLoRA)

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-hi for English → Hindi translation.
It was trained using QLoRA parameter-efficient fine-tuning on the CFILT IITB English-Hindi dataset.

The fine-tuned model shows improved translation quality compared to the base MarianMT model and it is faster than Base Model.

---

Model Details

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

• Model type: MarianMT (sequence-to-sequence translation model)
• Developed by: [Sagar / 54gO]
• Languages: English (en) → Hindi (hi)
• License: MIT
• Finetuned from: Helsinki-NLP/opus-mt-en-hi
• Dataset: cfilt/iitb-english-hindi
• Framework: 🤗 Transformers

Model Sources [optional]

• Repository: [More Information Needed]
• Paper [optional]: [More Information Needed]
• Demo [optional]: [More Information Needed]

Uses

Direct Use

• Translating English text into Hindi.
• Educational purposes (understanding fine-tuning of translation models).
• Research in low-resource language pairs.

[More Information Needed]

Downstream Use [optional]

• Integrate into chatbots for bilingual English-Hindi conversations.
• Use in document translation pipelines.
• As a teaching resource for NLP model fine-tuning with QLoRA.

Out-of-Scope Use

• Generating misinformation or harmful translations.
• High-stakes environments (medical, legal, safety-critical) without human verification.

---

[More Information Needed]

Bias, Risks, and Limitations

• May not handle rare domain-specific terms (e.g., medical/legal jargon) correctly.
• May introduce cultural biases or awkward phrasing.
• Performance depends on training dataset; Hindi dialects outside standard Hindi may not be well covered.

---

[More Information Needed]

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Use the code below to get started with the model.

How to Use

from transformers import MarianMTModel, MarianTokenizer

model_name = "YourUsername/marianmt-en-hi-translator"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)

text = "India is a beautiful country."
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs)
translation = tokenizer.decode(outputs[0], skip_special_tokens=True)

print(translation)
# भारत एक सुंदर देश है।

Training Details

Training dataset: cfilt/iitb-english-hindi

Fine-tuning method: QLoRA

Precision: 4-bit quantization with LoRA adapters

Batch size / Epochs: 16/10

Optimizer: AdamW

Scheduler: [cosine/linear/]

Compute: [T4 GPU, 30 minutes]

Metrics

BLEU: 81.43047985208148

CHRF: 87.07168424269955

TER: 16.29008746355685

Summary

✅ Faster inference compared to the base model, making it more practical for real-time applications.