repeated warning UserWarning: TypedStorage is deprecated

[hchau630]

🐛 Describe the bug

I noticed that the warning UserWarning: TypedStorage is deprecated that results from calling Tensor.storage() is issued repeatedly, even though the documentation about torch.set_warn_always says that by default pytorch warnings should only be issued once per process. While modifying my code to call untyped_storage() instead of storage() could get rid of the excessive warnings, I would prefer not to since I want to keep my code compatible with pre-2.0.0 versions for now, and untyped_storage() is only available >=2.0.0.

Example:

import torch

torch.tensor([1,2,3]).storage()
torch.tensor([1,2,3]).storage()
torch.tensor([1,2,3]).storage()

Output:

/path_to_file/test_warning.py:3: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
  torch.tensor([1,2,3]).storage()
/path_to_file/test_warning.py:4: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
  torch.tensor([1,2,3]).storage()
/path_to_file/test_warning.py:5: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
  torch.tensor([1,2,3]).storage()

Versions

Collecting environment information...
PyTorch version: 2.0.0
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 13.2.1 (arm64)
GCC version: Could not collect
Clang version: 12.0.0 (clang-1200.0.32.28)
CMake version: version 3.26.0
Libc version: N/A

Python version: 3.9.16 (main, Mar  1 2023, 12:19:04)  [Clang 14.0.6 ] (64-bit runtime)
Python platform: macOS-13.2.1-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Apple M1 Pro

Versions of relevant libraries:
[pip3] numpy==1.24.2
[pip3] torch==2.0.0
[conda] numpy                     1.24.2                   pypi_0    pypi
[conda] torch                     2.0.0                    pypi_0    pypi

cc @albanD

[albanD]

cc @kurtamohler could you fix that?

As a temporary workaround, you can silent this warning with regular python warning filter: python -W ignore::UserWarning: foo.py

[kurtamohler]

The complication is that the warning message contains the line of code that triggered the warning, with warnings.warn(..., stacklevel=...):

pytorch/torch/storage.py

Lines 365 to 372 in 0b094ca

	def _warn_typed_storage_removal(stacklevel=2):
	message = (
	"TypedStorage is deprecated. It will be removed in the future and "
	"UntypedStorage will be the only storage class. This should only matter "
	"to you if you are using storages directly. To access UntypedStorage "
	"directly, use tensor.untyped_storage() instead of tensor.storage()"
	)
	warnings.warn(message, UserWarning, stacklevel=stacklevel + 1)

So one option is to remove the stacklevel arg. But having the stack in the message is helpful for finding where the warning came from (#89867)

Another option is to only emit the warning the first time _warn_typed_storage_removal() gets called. For testing purposes, we could have a flag to make it always warn. Does that seem reasonable?

cc @ezyang

[ezyang]

Yeah, let's just have this warn only once and never again. If you're trying to scrub these, you can comment it out

[lorinczszabolcs]

Hi! Was this issue already solved? I tried to figure it out based on the PRs, but it seems none of them have been merged / released. Is there any suggested way of silencing it? Thanks!

[kurtamohler]

@lorinczszabolcs, #97379 has been merged to fix this issue, but it hasn't been included in a new release yet

Until it's released, you could use the workaround mentioned here: #97207 (comment)

But that will silence all UserWarnings. If you only want to silence the TypedStorage deprecation warnings, you'd have to do something like this:

import warnings
warnings.filterwarnings('ignore', category=UserWarning, message='TypedStorage is deprecated')

[RodolpheCalvet]

Hi there,
I was wondering about the implications/measures to take. Is it normal behavior, is pytorch gracefuly reverting to the new UntypedStorage class or not?
WHile running a streamlit web server locally I am good, but that's one of my currently possible reason for crash when deploying the app on streamlit.io.
Got torch==2.0.0 from pip...
Let me know if more info are relevant, tks:

[kurtamohler]

@RodolpheCalvet, it sounds like you're experiencing a crash related to UntypedStorage/TypedStorage. Could you please open a new issue and cc me on it? https://github.com/pytorch/pytorch/issues/new?assignees=&labels=&template=bug-report.yml