How to enable torch.profiler.profile using flag?

1

There is the context manager profile (PyTorch Profiler — PyTorch Tutorials 1.11.0+cu102 documentation) which allows us to monitor cpu and gpu times. I’d like to integrate it into my script such that I can turn it on / off flexibly:

from torch.profiler import profile
use_prof = True
with profile(enabled=use_prof,...): 
    # ...

Only there is no enabled option, as opposed to the torch.autograd.profile where such option exists (Automatic differentiation package - torch.autograd — PyTorch 1.11.0 documentation).

Is it possible / is there a workaround to flag the torch.profiler.profile somehow, just like torch.autograd.profile?

Thanks!

2

You can use a nullcontext and switch on that.

from torch.profiler import profile
import contextlib
use_prof = True
cm = profile(enabled=use_prof,...) if use_prof else contextlib.nullcontext()
with cm as prof:
   ...