Commit Graph

3 Commits

Author SHA1 Message Date
c68c38c673 Support getattr for tensor subclasses in pre-dispatch export via patching tensor.getattr (#143946)
Previous discussion: https://github.com/pytorch/pytorch/pull/143671#issuecomment-2560112499 and https://github.com/pytorch/pytorch/pull/143671

Differential Revision: [D67693609](https://our.internmc.facebook.com/intern/diff/D67693609)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143946
Approved by: https://github.com/bdhirsh
2025-01-06 23:55:50 +00:00
0ce233b8ca Support tensor subclass unwrapping (#141941)
This PR adds support for export to unwrap/wrap subclasses AOT so that we can trace through subclass parameters. This will resolve the UX issue in torchao where users had to manually unwrap their subclasses before calling export.

Differential Revision: [D67531057](https://our.internmc.facebook.com/intern/diff/D67531057)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/141941
Approved by: https://github.com/bdhirsh
2024-12-21 00:29:31 +00:00
6181e65cd8 Nested tensor subclass support (#127431)
When we have nested tensor subclasses, we need to recursively flatten/unflatten in Fake tensor creation and AOTAUtograd. Most of the PR is about mechanical change which changes today's single level flatten logic to be recursive.

Differential Revision: [D58533224](https://our.internmc.facebook.com/intern/diff/D58533224)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/127431
Approved by: https://github.com/bdhirsh
2024-06-26 04:45:22 +00:00