Find which line is using pytorch operation

autograd
1

I got this result from the profiler:

aten::ne 5.64% 372.377ms 16.59% 1.095s 39.548us 390.012ms 5.91% 1.099s 39.721us 27680

As far as I can tell through searching my code, I am not calling torch.ne anywhere. How can I trace exactly which lines are using this function?

2

Probably some other function is calling torch.ne, i.e. comparing if tensor elements are smaller than or equal elementwise…

3

You could check if the profiler shows this method in its stacktrace as shown here and might then be able to narrow it down.