Incorrect API call

[shashank-shivam]

System Info

torch.get_default_device() is not an official PyTorch API (even in 2.2), but it’s being called inside modeling_utils.py in version 4.52.1.

get_torch_context_manager_or_global_device() function is calling the following API
Error - Exception : module 'torch' has no attribute 'get_default_device'

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

HuggingFaceEmbeddings(model_name=self._model_name)

Initialization fails with Exception : module 'torch' has no attribute 'get_default_device'

Expected behavior

Can be replaced with
default_device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

[mahimairaja]

@shashank-shivam Can you please add some relevant details

• which platform
• python version
• pytorch version

[Rocketknight1]

Hi @shashank-shivam, torch.get_default_device is part of the PyTorch API. Have you tried updating your version of PyTorch?

[shashank-shivam]

Standard PyPI repository doesn't have new releases of torch
"ERROR: Could not find a version that satisfies the requirement torch==2.7.0 (from versions: 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1, 2.0.0, 2.0.1, 2.1.0, 2.1.1, 2.1.2, 2.2.0, 2.2.1, 2.2.2)"

For more context:
Platform macOS 15.4.1
Python 3.10
pytorch 2.2.2

[Rocketknight1]

Hi @shashank-shivam, is it possible you have an x86 mac? Newer versions of Torch are probably not getting released there and Apply is dropping support?

[shashank-shivam]

Yes that's correct. Torch doesn't seem to release newer version after 2.2.2 for mac x86intel. But I was wondering that the code can overcome the hardware dependency of libraries specially if it doesn't have a core functionality dependency. But, that's just my thought. Feel free to close the bug if that doesn't make sense

[Rocketknight1]

Hi @shashank-shivam yes, good point. I think the latest version on main doesn't use get_default_device anymore so torch 2.2.2 should continue working for another while