distribution of performance

[sinarafati-amd]

for estimating the inference time instead of running it once measure variance across multiple runs to have better stats