Hardware-Aware Attention Design
Designs attention layers around accelerator hardware constraints to speed up training or inference without losing model quality.
2 papers
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.