torch.export non strict mode uses "make_fx [which] uses __torch_dispatch__ to trace under the hood. this is where it creates the fx nodes.
AOTAutograd also calls into make_fx, but before that it also does some things related to functionalization.
Since export is now "Training IR", it no longer does functionalization, so we just directly call make_fx."
Created
August 13, 2025 18:44
-
-
Save justinchuby/f39ca941efcbd8e826cac57657d5dda9 to your computer and use it in GitHub Desktop.
PyTorch export non-strict vs strict modes
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment