Worked on the AMD-AGI/Primus repository to enhance configuration flexibility in machine learning workflows. Developed and integrated a new wandb_enable configuration option within the Torchtitan examples, allowing users to toggle Weights & Biases logging as needed. The implementation focused on robust configuration management using Python and YAML, ensuring seamless integration with existing machine learning frameworks. To guarantee reliability, comprehensive unit tests were written to validate correct parsing and handling of the new configuration option. This work improved the adaptability of experiment tracking in Torchtitan, supporting more controlled and customizable logging for users without introducing any new bugs during the development period.