torch.jit.trace() fix

[glenn-jocher]

Signed-off-by: Glenn Jocher glenn.jocher@ultralytics.com

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhancement of model export functionality and CI testing for YOLOv5 🚀.

📊 Key Changes

• Added warmup and tracing to the Continuous Integration (CI) testing workflow for better simulation of real-world usage.
• Included an extra image processing step in CI testing to ensure model functionality.
• Updated the models/common.py file to prevent the outputting of loss values during export.

🎯 Purpose & Impact

• 🎭 Improves testing robustness: The warm-up process and inclusion of the JIT trace ensure that CI testing resembles actual conditions more closely, leading to more reliable code.
• 🧠 Models are easier to deploy: Suppressing loss output during model export makes the models cleaner for inference purposes, which is beneficial for users who want to deploy models easily into production environments.
• 🛠 Potentially reduces issues: These changes could lead to catching potential errors earlier, improving the overall quality of the YOLOv5 codebase.