Deprecate variadic inputs of checkpoint_sequential #21006
Conversation
pytorchbot
added
module: activation checkpointing
|
To test I fixed this problem by modifying |
torch/utils/checkpoint.py
Outdated
| # This interface will be changed at PyTorch 1.2.0. | ||
| # See also: https://github.com/pytorch/pytorch/issues/19260 | ||
| if not inputs: | ||
| warnings.warn('no input to checkpoint_sequential has been deprecated, ' |
There was a problem hiding this comment.
"no input" -> "giving no inputs"
"will be raised since PyTorch 1.2.0" -> "will be raised since PyTorch 1.3.0" (we need atleast 1 stable version to deprecate things, right now it is 1.1.0)
There was a problem hiding this comment.
Great feedback and fix. I also capitalized the warning message for multiple inputs just like "Giving no inputs". Thank you so much!
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@soumith is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Summary: To support variadic inputs of `checkpoint_sequential` was deprecated at #21006. This case should be warned with `DeprecationWarning` for PyTorch 1.2, but it should be simply failed with `TypeError` since PyTorch 1.3. This patch removes the `DeprecationWarning` for PyTorch 1.2. Pull Request resolved: #25985 Differential Revision: D18809875 Pulled By: albanD fbshipit-source-id: e84dd8629c04979c4b2dc63e8ada94292e8cedd0
Summary: To support variadic inputs of `checkpoint_sequential` was deprecated at pytorch#21006. This case should be warned with `DeprecationWarning` for PyTorch 1.2, but it should be simply failed with `TypeError` since PyTorch 1.3. This patch removes the `DeprecationWarning` for PyTorch 1.2. Pull Request resolved: pytorch#25985 Differential Revision: D18809875 Pulled By: albanD fbshipit-source-id: e84dd8629c04979c4b2dc63e8ada94292e8cedd0
6 participants
I've reported inconsistency between
checkpoint_sequentialandnn.Sequentialat #19260. Both should provide the same input signature but they don't. I think the consistency is important and I agree with @apaszke thatnn.Sequential's semantics should be kept instead ofcheckpoint_sequential.I hope
checkpoint_sequentialraisesTypeErroron variadic arguments since PyTorch 1.2.0. But for now, it's okay just to warn asDeprecationWarning. I've talked about this approach with @soumith.Please review this pull request. Any comment will be my pleasure.