English-to-Nepali-translation

If you want to play around with the notebook, click the following button.

About the code

This repository contains code for translating English text into Nepali using the M2M100 model from Hugging Face's Transformers library. The model is fine-tuned for sequence-to-sequence (seq2seq) translation tasks.

Requirements

Python 3.10
PyTorch
Transformers
Datasets

Code overview

English_to_Nepali_translation.ipynb: Contains code for training the seq2seq model and performing English to Nepali translations.

Example

english_text = "I am going to the store to buy some groceries."

# Tokenize the input English text using the tokenizer
inputs = tokenizer(english_text, return_tensors='pt')

# Generate the Nepali translation of the input text using the trained model
generated_tokens = model.generate(inputs['input_ids'].to("cuda"),
                                  attention_mask=inputs['attention_mask'].to("cuda"),
                                  num_beams=4,
                                  max_length=512)

# Decode the generated tokens to get the final Nepali translation
nepali_translation = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
print("Nepali Translation:", nepali_translation)

Acknowledgement

The code uses the M2M100 model by Alirezamsh from the Hugging Face Model Hub.
The code for computing BLEU score utilizes the sacreBLEU metric.