smart_optimizer() revert to weight with decay

[glenn-jocher]

If a parameter does not fall into any other category

Signed-off-by: Glenn Jocher [email protected]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced parameter grouping in optimizer setup for YOLOv5 models.

📊 Key Changes

• Refactored the way parameters are added to optimizer groups, simplifying the code.
• Switched to using named_parameters method to iterate over parameters, which is a more general approach.
• Parameters are now grouped based on their name rather than just their type, allowing for potential custom-named parameters to be correctly categorized for optimization.

🎯 Purpose & Impact

• Purpose: To ensure that all model parameters are correctly grouped for optimization, with or without weight decay, based on their role in the model (e.g., weights or biases).
• Impact: This change leads to more ensure comprehensive optimization of all model parameters, potentially improving model training efficiency and accuracy. This could benefit users by providing better-performing models with possibly faster convergence times.