Filter PyTorch 1.13 UserWarnings

[triple-Mu]

When training a model, I get such warnings.

Maybe we should filter pytorch 1.13 warning about torch.distributed._all_gather_base or replace it with torch.distributed.all_gather_into_tensor

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved warning management for CUDA availability in PyTorch.

📊 Key Changes

• Added suppression of general UserWarning category warnings in the torch_utils.py file.

🎯 Purpose & Impact

• 🛠 Purpose: To reduce user confusion by suppressing non-critical warnings that might not directly affect the user experience or functionality of the software.
• 🧑‍💻 Impact: Users will encounter a cleaner console output, with fewer distractions from warnings that do not require immediate action, enhancing user experience.

[glenn-jocher]

@triple-Mu thanks for the PR! Yes we should probably filter this message. I started seeing another message also with DDP just like this user:
#10150

UserWarning: is_namedtuple is deprecated, please use the python checks instead

[glenn-jocher]

@triple-Mu we don't use torch.distributed._all_gather_base in YOLOv5 directly, I think torch 1.13 is using it internally and causing this warning, and same with is_namedtuple. I guess this PR will resolve both of these warnings simultaneously?

[triple-Mu]

@triple-Mu we don't use torch.distributed._all_gather_base in YOLOv5 directly, I think torch 1.13 is using it internally and causing this warning, and same with is_namedtuple. I guess this PR will resolve both of these warnings simultaneously?

It will not only filter torch warning but also other python userwarning.
BTW Logger.warning not affected

[glenn-jocher]

@triple-Mu PR is merged. Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐