Open caddyless opened 2 years ago
Hi @caddyless,
Thanks for your comment. Actually, I didn't observe notable differences in the results of the throughput testing with/without torch.no_grad()
. The differences were all within 1%, so there was no need to add torch.no_grad()
. Did you find a large difference in the throughput testing with/without torch.no_grad()
?
I found torch.no_grad() is missed in the speed_test.py, which significantly slows down the throughput. Is this intentional or a mistake?