bash run.sh using m=10000000 n=64 CUDA kernel takes 10 ms verified Input shape torch.Size([10000000, 64]) TORCH max takes 24.503946 ms Tested on sm70 and sm86(RTX 4060Ti)