Transformer implementation adapted from the tensorflow example. Models trained on the Portuguese to English ted datasets, the same as the tensorflow example.

A small and large model are trained, as detailed by configs files. For both, val_accuracy reaches a roughly 0.35 maximum. BLEU and GLEU scores are calculated on the validation set after training using a few saved checkpoints. See the inference.ipynb notebook for results.