yolo.py profiling updates

[glenn-jocher]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced model profiling options in YOLOv5 for better performance analysis.

📊 Key Changes

• 🧪 Introduced --batch-size argument to specify batch size for profiling.
• 📈 Added --line-profile flag to enable detailed layer-by-layer model profiling.
• 🔍 Refined the profiling process by utilizing the profile function of torch_utils.
• 🚀 Modified profiling to support forward and backward pass performance analysis.
• 🧹 Removed old Tensorboard code that was not working properly.

🎯 Purpose & Impact

• 📐 Allows users to fine-tune the profiling batch size for their specific hardware resources, optimizing the profiling process.
• 🛠️ Enables developers to identify performance bottlenecks at the per-layer level, which is critical for targeted model optimization.
• 🏎️ Gives the option to profile the model's forward and backward passes to assess overall computational performance.
• 🧽 Codebase cleanup increases maintainability and focuses on functional profiling tools.
• 🤖 Enhanced profiling capabilities can lead to better optimized YOLOv5 models, improving speed and efficiency for all users.