Attention Is All You Need

The paper that introduced the Transformer architecture — covering scaled dot-product attention, multi-head attention, positional encodings, and the original encoder-decoder design.