pytorch

f6f46ba3 - [Reland] aot autograd explicitly errors on double backward (#92893)

[Reland] aot autograd explicitly errors on double backward (#92893) This reverts commit fb980581a7b41a5ea570fcb03829463b806b3bbc. Testing: `python benchmarks/dynamo/timm_models.py --float32 --training --only=mobilevit_s --performance --inductor --disable-cudagraphs` ``` main: memory: eager: 12.30 GB, dynamo: 12.28 GB, ratio: 1.00 + #90896 reverted: memory: eager: 12.30 GB, dynamo: 8.81 GB, ratio: 1.40 + this PR: memory: eager: 12.30 GB, dynamo: 8.81 GB, ratio: 1.40 ``` For comparison, if we apply old version of this PR instead: ``` main: + #90896 reverted: memory: eager: 12.30 GB, dynamo: 8.81 GB, ratio: 1.40 + old version of this PR memory: eager: 12.30 GB, dynamo: 10.36 GB, ratio: 1.19 ``` Pull Request resolved: https://github.com/pytorch/pytorch/pull/92893 Approved by: https://github.com/bdhirsh