自注意力

Attention computed between all pairs of elements within one sequence.

Self-attention lets every position directly "see" every other position, capturing word-word and sentence-sentence relations—key to understanding context. Its cost grows quadratically with sequence length, the main bottleneck for long contexts.