Autotune not working with x86 AVX2 architecture

[kevinthesun]

@e0lithic

the new compiled models are having slower inference time

Are you comparing with default schedule?