MarianMT Fine-tuned on Tatoeba (en → fr)

This repository contains a Transformer-based Neural Machine Translation (NMT) pipeline where a pretrained MarianMT model was fine-tuned on the Tatoeba English–French parallel corpus.

Languages

• English → French

Base Model

• MarianMT: Helsinki-NLP/opus-mt-en-fr

Dataset

• Tatoeba EN–FR parallel corpus: Helsinki-NLP/tatoeba

Metrics

Model	BLEU Score
Base OPUS-MT (MarianMT)	50.5
Fine-tuned OPUS-MT	55.44

Repository Structure

• finetuned-opus-en2fr.ipynb — training + evaluation pipeline
• demo.ipynb — sample translations
• demo.py — streamlit demo