Skip to content

other

Multi-Query Attention

Attention design in which every query head shares a single key-value head, shrinking cached state at the cost of representational capacity.

Known aliases

  • MQA

Relationships

No evidence-backed relationships are recorded.

Current clusters