... and it's different flavours
- simplified self-attention
self-attentioncausal attentionmulti-head attention
- encoder-decoder to account for different order of input and output sequences
contextis important, because it creates enriched representations of each element in an input sequence, by incorporating relevant information from all other elements in the sequence
- weight parameters vs attention weights
W_query,W_key,W_value- optimized during training (like weights query, etc.)attn_weights- computed on the fly for each input sequence (like attention weights)

