ignite/handlers/checkpoint.py:16: DeprecationWarning concerning TorchScript

[sjiang95]

🐛 Bug description

ignite logger triggers the following warning everytime (maybe) a new logger instance is created.

ignite/handlers/checkpoint.py:16: DeprecationWarning: `TorchScript` support for functional optimizers is deprecated and will be removed in a future PyTorch release. Consider using the `torch.compile` optimizer instead.
  from torch.distributed.optim import ZeroRedundancyOptimizer

Reproduce

Below is a snippet to reproduce it

import torch
import ignite


def foo(i, logger):
    pass


if __name__ == "__main__":
    logger = ignite.utils.setup_logger()
    torch.multiprocessing.spawn(foo, args=(logger,), nprocs=2)

stdout:

/opt/conda/lib/python3.11/site-packages/ignite/handlers/checkpoint.py:16: DeprecationWarning: `TorchScript` support for functional optimizers is deprecated and will be removed in a future PyTorch release. Consider using the `torch.compile` optimizer instead.
  from torch.distributed.optim import ZeroRedundancyOptimizer
/opt/conda/lib/python3.11/site-packages/ignite/handlers/checkpoint.py:16: DeprecationWarning: `TorchScript` support for functional optimizers is deprecated and will be removed in a future PyTorch release. Consider using the `torch.compile` optimizer instead.
  from torch.distributed.optim import ZeroRedundancyOptimizer
/opt/conda/lib/python3.11/site-packages/ignite/handlers/checkpoint.py:16: DeprecationWarning: `TorchScript` support for functional optimizers is deprecated and will be removed in a future PyTorch release. Consider using the `torch.compile` optimizer instead.
  from torch.distributed.optim import ZeroRedundancyOptimizer

where the first warning is triggered by ignite.utils.setup_logger(), and the rest nprocs times warnings are triggered when the sub processes are initialized.

After some trials, I found that the warning is triggered once ignite.handlers.checkpoint is imported, for which simply importing ignite also leads to the warning.

$ python -c "import ignite"
/opt/conda/lib/python3.11/site-packages/ignite/handlers/checkpoint.py:16: DeprecationWarning: `TorchScript` support for functional optimizers is deprecated and will be removed in a future PyTorch release. Consider using the `torch.compile` optimizer instead.
  from torch.distributed.optim import ZeroRedundancyOptimizer
$ python -c "import ignite.handlers"
/opt/conda/lib/python3.11/site-packages/ignite/handlers/checkpoint.py:16: DeprecationWarning: `TorchScript` support for functional optimizers is deprecated and will be removed in a future PyTorch release. Consider using the `torch.compile` optimizer instead.
  from torch.distributed.optim import ZeroRedundancyOptimizer

Expect

Avoid the annoying warning message.

Environment

• PyTorch Version: 2.5.1_cuda12.4_cudnn9_0
• Ignite Version: 0.5.1
• OS: Linux and Windows
• How you installed Ignite: conda install -c pytorch ignite
• Python version: 3.12, 3.11
• Any other relevant information:

relevant issues

#3318
Project-MONAI/MONAI#8060
Project-MONAI/MONAI#8017 (comment)

[sjiang95]

This is fixed by
pytorch/pytorch#140889.

[vfdev-5]

Thanks for digging into the issue @sjiang95 !
Hopefully, new version pytorch wont raise this warning anymore. Nightly version should be silent, to confirm...

[sjiang95]

I have confirmed the issue has been solved in pytorch nightly build.
Test env is
pytorch: 2.7.0.dev20250108+cu126
ignite: 0.5.1