Background: I am profiling my model to see the bottleneck. I am not interested in the dataloader, so ignoring its performance is ok for me.
I use the profiler as shown below:
with torch.autograd.profiler.profile(
enabled=True,
use_cuda=False,
record_shapes=True,
with_flops=True,
profile_memory=True,
use_cpu=True
) as prof:
with torch.no_grad():
for _ in range(args.nepochs):
for i, test_batch in enumerate(test_ld):
X_test, y_test, z_test, T_test = test_batch
R_test = model_wrap(
model,
X_test,
y_test,
z_test,
)
And later I found the enumerate(DataLoader)#_SingleProcessDataLoaderIter._... occupies more than 90% in the breakdown, though I don’t care about it. It makes the log unreadable. How can I remove its information from the profiler?