2 Commits

Author	SHA1	Message	Date
Sherlock Huang	a956c4ab1c	Return NoOpDeviceGuardImpl in replace of CudaDeviceGuard when device is not available, or cpu-only build (#160532) ... Summary: To support exporting a cuda model on a CPU-only machine under fake tensor mode. User commonly need to move sample inputs to the cuda device with .to("cuda:0") or .to("cuda") call. This diff supports this. I expect the following pattern to work ``` with FakeTensorMode(allow_non_fake_inputs=True): cuda_module = module.to("cuda:0") cuda_sample_inputs = tuple([x.to("cuda:0") for x in sample_inputs]) with torch.no_grad(): ep = torch.export.export(cuda_module, cuda_sample_inputs) ``` Test Plan: CI Rollback Plan: Differential Revision: D80181887 Pull Request resolved: https://github.com/pytorch/pytorch/pull/160532 Approved by: https://github.com/henryoier, https://github.com/ezyang	2025-09-13 01:50:51 +00:00
Sherlock Huang	eaa5d9d3d3	Introduce OpInfo test for testing export on fake device (#160694) ... Summary: Prepare for the upcoming diffs for exporting on fake cuda device. Test Plan: test Rollback Plan: Differential Revision: D80304225 Pull Request resolved: https://github.com/pytorch/pytorch/pull/160694 Approved by: https://github.com/dolpm	2025-08-15 07:26:28 +00:00