Can PyTorch1.10 cuda113 profiling show op's name in aten levels


Can PyTorch 1.10 profiling show aten level ops’ name ?

When we previously used PyTorch 1.7, the profiling was very useful and we can see aten level ops’ performance according to its name.

While i use PyTorch 1.10, the profiling info is shown as below:

With this name, i can not know which is conv and which is batchnorm. Could we profile like PyTorch1.7 ?

Thank you. :slight_smile: