Post by @leading • Hey

Inference cost comparison of Transformer and ResNet at 6.7B model size. ResNet outperforms clearly