Multi-Head Attention

Stores past attention states to speed up autoregressive decoding.

Encodes positional information via rotation in embedding space.

Encodes token position explicitly, often via sinusoids.

Techniques to handle longer documents without quadratic cost.

The range of functions a model can represent.

Tradeoffs between many layers vs many neurons per layer.

Prevents attention to future tokens during training/inference.

A single attention mechanism within multi-head attention.