This is an English to German translator built using Transformers. The transformer was implemented from scratch using TensorFlow 2 and the instructions of the paper "Attention Is All You Need"
The data can be found in:
https://www.statmt.org/europarl/
And the non_breaking_prefixes can be found in this repository:
https://github.com/moses-smt/mosesdecoder/tree/master/scripts/share/nonbreaking_prefixes
I did not have a proper GPU so I used google colab (not pro) for the training. I chose a very very smaller model compared to the paper and I trained it only for 2 epochs. After one hour of training, the model achieved 24% accuracy at the end of the first epoch and 36% on the second, this could easily be further improved with more epochs and a larger model and more parameter tuning which I don't have the proper GPU for but the results are pretty good considering the small model size and epochs.
This translator is inspired from this udemy course: https://www.udemy.com/course/modern-nlp/