How to map lowered Function with Task Name in AutoTVM?


#1

Hi,

I am wondering where to find this mapping? I have a resnet-50 model. After compilation, the graph has 58 fused functions but only 20 layers of convolution for tuning. So, where could I find the mapping? The purpose is to selectively tuning layers that takes more time.

Thanks.


#2

You can print the workload of each conv2d: tsk.workload.