Unified-Normalization
Unified-Normalization copied to clipboard
about speed up
Firstly,thinks for your contributions to the transformer deployment work. I focus on a 31% speed up increase, Please, Whether this performence is based on pytorch or TensorRT? By the way, Does Swin_Transformer model get 31% performence increse in speed boost by replacing the default layer-normalization with unfied-normalizaiton with TensorRT environment?