ConceptAI & ML
Sliding window attention
An attention mechanism where each token only attends to a local neighborhood of surrounding tokens to increase efficiency.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Lecturers
2Each in their own terms, from the easiest level.
Albert Jiang
An attention variant where tokens in lower transformer layers attend only to a short local span of previous tokens, accumulating broader context in deeper layers.
Connections
4The lectures that define it or assume it, and the concepts next to it.