rzou 82d00acfee Allow cross-device copies for cpu scalars in refs (#135140) ...

This copies our eager-mode behavior where someone can do torch.add(a, b, out=c)
where a and b are CPU scalar tensors and c is a CUDA tensor.

Fixes https://github.com/pytorch/pytorch/issues/121619 by side effect (we get into a situation where we're writing a CPU scalar into a FakeTensor that is actually a meta tensor)

Test Plan:
- new test

Pull Request resolved: https://github.com/pytorch/pytorch/pull/135140
Approved by: https://github.com/williamwen42, https://github.com/yanboliang

2024-09-05 19:08:48 +00:00

..

__init__.py

Remove unused imported names in python files (#134438)

2024-08-27 20:44:04 +00:00

wrappers.py

Allow cross-device copies for cpu scalars in refs (#135140)

2024-09-05 19:08:48 +00:00