W&B artifacts feature addition

[glenn-jocher]

This PR adds W&B artifact upload ability with the --log-artifacts flag in train.py. When this flag is used final trained model (best.pt if available, else last.pt if --nosave is used) will be uploaded to the W&B cloud console.

$ python train.py --log-artifacts

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced model saving and logging for YOLOv5 training.

📊 Key Changes

• Simplified decision logic for choosing between 'best' or 'last' saved models.
• Centralized uploading of the final model to a storage bucket (e.g., Google Cloud Storage).
• Added conditional logging of model artifacts on Weights & Biases (W&B).
• Ensured the final model is used for post-training evaluation.

🎯 Purpose & Impact

• Code Clarity: Condensed and cleaned up the code dealing with model saving.
• Efficiency: Avoids redundant uploads by choosing only the final model for cloud storage.
• Integration: Provides better integration with W&B by offering artifact logging.
• Consistency: Guarantees the usage of the same final model for subsequent evaluation, ensuring consistent results.

The update streamlines the training end process, making it clearer, more efficient, and better integrated with external tools like W&B, enhancing the user experience for developers leveraging the YOLOv5 training pipeline. 🚀

[glenn-jocher]

@AyushExel I've created an artifacts upload feature addition with W&B!

[AyushExel]

@glenn-jocher looks good!