Time profiler

[Jeffrey-Joan]

• Model Loading: Loads a YOLO model from a specified path using the Ultralytics library.
• Random Input Tensor Creation: Creates a random input tensor with user-specified batch size, height, and width.
• CUDA Support: Optionally moves the model and input tensor to CUDA if available, enhancing profiling performance.
• Profiling Model Execution: Uses PyTorch's autograd.profiler to profile the model's execution and provides detailed performance metrics.
• Profiling Results Output: Prints the profiling results to the console, offering insights into the model's execution time, memory usage, and other performance indicators.