Fine-tuned mBART Model for Urdu to English Translation
This repository contains a fine-tuned mBART model for Urdu to English translation. The model has been trained on a custom dataset and evaluated on test data.
Model Information
- Model Name:
abdulwaheed1/urdu_to_english_translation_mbart - Base Model:
facebook/mbart-large-50 - Tokenizer:
facebook/mbart-large-50 - Source Language: Urdu (
ur) - Target Language: English (
en)
Usage
python
from transformers import MBart50TokenizerFast, MBartForConditionalGeneration
# Load the fine-tuned model
model_name = "abdulwaheed1/urdu_to_english_translation_mbart"
tokenizer = MBart50TokenizerFast.from_pretrained(model_name, src_lang="ur_PK", tgt_lang="en_XX")
model = MBartForConditionalGeneration.from_pretrained(model_name)
Evaluation
The model has been evaluated on a test dataset, and the following metrics were obtained:
- BLEU Score: 30.95
- Generation Length: 35.90
- Meteor Score: 0.58
Training Details
The model was trained using the transformers library with the following configuration:
- Training Loss: 1.450300
- Validation Loss: 1.281330
Dataset
The model was fine-tuned on a custom Urdu-English translation dataset. If you wish to use the same dataset, you can find the preprocessing script and dataset files in the data directory.
Acknowledgments
The fine-tuning process and code were inspired by the Hugging Face Transformers library.
- Downloads last month
- 108