Multi-GPU default to single device 0

[glenn-jocher]

Default to device 0 on Multi-GPU systems. This may help prevent instances of DP training when DDP commands should be used for best results. See Multi-GPU Tutorial for full details.

WARNING: BREAKING WORKFLOW CHANGES

• Previously: all available GPUs were used on Multi-GPU systems if no --device was passed.
• This update: device 0 will be used by default on Multi-GPU systems if no --device. You now need to opt-in to any GPU groupings using the --device argument, i.e. python train.py --device 0,1,2,3

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Simplified default GPU selection in YOLOv5's device utility.

📊 Key Changes

• Changed the default GPU device from a range of all available GPUs to just '0' (the first GPU).
• Adjusted the padding space for better alignment in the printed output.

🎯 Purpose & Impact

• Ensures a single GPU is used by default, preventing possible unintended use of multiple GPUs when the device is unspecified.
• Improves the readability of the output by neatly aligning the CUDA device information.
• This change will mostly affect users who do not specify the GPUs they want to use; it ensures the script doesn't accidentally occupy all available GPUs, which is important for shared systems.

[glenn-jocher]

/rebase

[glenn-jocher]

Verification example: