The Transformer (self-attention)

Step through Query, Key and Value projections, attention weights, residual connections and token prediction. Every word looks directly at every other word, in parallel.

Reference Vaswani et al., 2017 — “Attention Is All You Need”