This model is fine-tuned on 500 Roman → Nepali sentence pairs using the facebook/mbart-large-50 base model.
It is intended for educational and fun NLP projects, allowing conversion of Romanized Nepali text into proper Nepali script.

Developed by: rajeshrai
Model type: MBart50
Language(s): Romanized Nepali → Nepali (ne_NP)
License: MIT (or your preferred license)
Fine-tuned from: facebook/mbart-large-50

Model Sources

Base Model Repository: facebook/mbart-large-50

Uses

Direct Use

Translate Romanized Nepali text to Nepali (Devanagari) for educational purposes, chatbots, or small-scale projects.

Out-of-Scope Use

This model is not suitable for professional translation or large-scale production as it is trained on only 500 sentence pairs.
It may produce inaccurate translations for complex or long sentences.

Bias, Risks, and Limitations

Trained on a small dataset, so coverage is limited.
Might struggle with slang, rare words, or long sentences.
Outputs should be reviewed by a human for critical use.

How to Get Started

python

from transformers import MBartForConditionalGeneration, MBart50TokenizerFast

model_name = "rajeshrai577/roman-to-nepali-mbart50"

model = MBartForConditionalGeneration.from_pretrained(model_name)
tokenizer = MBart50TokenizerFast.from_pretrained(model_name, src_lang="en_XX", tgt_lang="ne_NP")

input_text = "translate Roman Nepali to Nepali: ma ghar jaanchu."
inputs = tokenizer(input_text, return_tensors="pt")
generated_ids = model.generate(
    **inputs,
    forced_bos_token_id=tokenizer.lang_code_to_id["ne_NP"],
    max_length=50
)
print(tokenizer.decode(generated_ids[0], skip_special_tokens=True))
# Output: म घर जान्छु।