May 27, 2022
An IO-aware attention algorithm reduces memory traffic while computing exact attention.
Open implementations help make longer-context model training and inference more efficient.
Explore on Humanity Unfolds
