Basic implementation of BERT and Transformer in Pytorch in one python file of ~300 lines of code.
This project aims to provide an easy-to-run easy-to-understand code for NLP beginners and people who want to know how Transformers work.
The project uses a simplified implementation of BERT (unsupervised learning).
The original implementation of Transformer uses an encoder and a decoder, here we only need the encoder.
The model can train in 30 minutes on 1 x RTX2070Super GPU.
Visualization of word embeddings:
Implementation details: https://hyugen-ai.medium.com/transformers-in-pytorch-from-scratch-for-nlp-beginners-ff3b3d922ef7