attempt_load() deserialize fix

[glenn-jocher]

Per #8001 (comment)

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhancement of the model loading mechanism in the YOLOv5 library.

📊 Key Changes

• Modified the attempt_load function within experimental.py.
• Streamlined model weight loading by combining torch.load with map_location.

🎯 Purpose & Impact

• Purpose: Simplifies the process of loading the model's weights onto the specified device, ensuring that the weights are immediately placed on the correct device during the loading process.
• Impact: Improves efficiency by reducing device-to-device memory copy operations which can speed up initialization times, especially for systems with multiple GPUs or CPU-only systems. This can enhance the user experience by providing a more seamless setup and potentially reducing bugs related to device-specific weight loading.