Basic module for caveman-style PyTorch memory usage debugging
Are you trying to run PyTorch models without access to a supercomputer? Is your model using 28 yottabytes of memory when loading?
This module won't help you with either of those, but it might point fingers at what it thinks is eating memory.
import tikitrace
@tikitrace.trace()
def do_something_crazy():
torch.load("blah blah blah")crazy right? it gets even crazier:
import tikitrace
with tikitrace.trace_ctx():
torch.load("blah blah blah")you're probably right, file an issue and ill fix it
you're probably right, file an issue and ill fix it