Transformer Network Trained With Back-Translation

Assumes python3 is being used.

To train:

python train.py [PARAMS]

To evaluate:

python evaluate.py --model [PATH TO MODEL TO EVALUATE]

To translate:

Some configurations to translate.py are required to translate custom sentences. translate.py is currently set up to spit out DE->EN translations of the IWSLT test split as provided by TorchText.

To preprocess:

preprocess.py sets up the translations from translate.py to be importable by TorchText.

Model Weights:

Download the weights from here

https://drive.google.com/file/d/1AGG5sdyEPr9uCSv-1KXnFkIk-qYnlClW/view?usp=sharing

and extract them into models/ directory.