Seq2Seq Translation Model (English to Norwegian)
Overview
This project implements a Sequence-to-Sequence (Seq2Seq) model to translate sentences from English to Norwegian. Using PyTorch and the OPUS100 dataset, the model leverages LSTM-based encoder-decoder architecture to generate translations.
Highlights
Dataset: English-Norwegian language pairs from OPUS100.
Model:
Embedding layer for word representation.
LSTM-based encoder-decoder architecture.
Linear layer for output predictions.
Metrics:
BLEU and ChrF scores to evaluate translation quality.
Visualizations:
Loss curves and evaluation scores plotted over training epochs.
Workflow
Data Preprocessing:
Tokenization and padding of sequences for uniform input.
Model Training:
CrossEntropyLoss and Adam optimizer for efficient learning.
Evaluation:
BLEU and ChrF scores to measure performance.
Model Saving:
Trained weights stored as seq2seq_model.pth.
Visualization:
Loss trends and evaluation metrics visualized for insights.
