Transformer

Coming soon...

Unofficial code for the original paper Attention is all you need by Ashish Vaswani et al.

TODO

• Sublayer Residual Connection
• Encoder Layer
• Encoder
• Decoder Layer
• Decoder
• Multi-Head Attention
• Position-wise Fully Connected Feed-Forward Network
• Positional Encoding
• Embedding
• PyTorch Transformer
• Training Data
• Utils
• Training Scripts
• Train
• JAX implementation

Acknowledgements

I was heavily inspired by

• The original paper Attention Is All You Need by Ashish Vaswani et al., see here
• OpenNMT: Open-Source Toolkit for Neural Machine Translation by Guillaume Klein et al., see here
• Illustrated Attention by Raimi Karim, see here
• pytorch-original-transformer by Aleksa Gordić, see here

License