Simplify _NO_WRAPPING_EXCEPTIONS #7806
Conversation
| @@ -203,4 +203,3 @@ def test_deepcopy(datapoint, requires_grad): | |||
|
|
|||
| assert type(datapoint_deepcopied) is type(datapoint) | |||
| assert datapoint_deepcopied.requires_grad is requires_grad | |||
| assert datapoint_deepcopied.is_leaf | |||
There was a problem hiding this comment.
the deepcopy isn't a leaf anymore because it went through wrap_like(), so it's got an "ancestor".
I don't think is_leaf is part of the deepcopy contract anyway? I don't think we really need to enforce this.
There was a problem hiding this comment.
the deepcopy isn't a leaf anymore because it went through
wrap_like(), so it's got an "ancestor". I don't thinkis_leafis part of thedeepcopycontract anyway? I don't think we really need to enforce this.
I don't think it is specified anywhere, so I'm ok with removing this check. Might be surprising to users though if they bank on this. Let's find out though 🤷
| if isinstance(output, cls): | ||
| # DisableTorchFunctionSubclass is ignored by inplace ops like `.add_(...)`, | ||
| # so for those, the output is still a Datapoint. Thus, we need to manually unwrap. |
There was a problem hiding this comment.
I also move most of the content of the DisableTorchFunctionSubclass out of it. The only part that matters is the call to func, the rest can be outside.
| @@ -203,4 +203,3 @@ def test_deepcopy(datapoint, requires_grad): | |||
|
|
|||
| assert type(datapoint_deepcopied) is type(datapoint) | |||
| assert datapoint_deepcopied.requires_grad is requires_grad | |||
| assert datapoint_deepcopied.is_leaf | |||
There was a problem hiding this comment.
the deepcopy isn't a leaf anymore because it went through
wrap_like(), so it's got an "ancestor". I don't thinkis_leafis part of thedeepcopycontract anyway? I don't think we really need to enforce this.
I don't think it is specified anywhere, so I'm ok with removing this check. Might be surprising to users though if they bank on this. Let's find out though 🤷
|
Hey @NicolasHug! You merged this PR, but no labels were added. The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py |
Reviewed By: matteobettini Differential Revision: D48642278 fbshipit-source-id: b74f5744ca32672d70f89dab2e8ef01b073c3be0
by calling
wrap_likefor all the ops in_NO_WRAPPING_EXCEPTIONSinsteaf of having a custom logic for each one.We don't need to save a few us on calls to
requires_grad_().cc @vfdev-5