Worked on the lab-cosmo/pet-mad repository to enhance the reliability of MADExplorer’s initialization process. Focused on stabilizing device management by centralizing how the pet model is assigned to computational devices, this update addressed device-mismatch errors that previously affected multi-device experiments. The solution involved moving the pet instance to the correct target device during initialization, ensuring consistent behavior across different hardware setups. Using Python and PyTorch, the developer delivered a well-documented fix that improved both traceability and maintainability. This work laid the foundation for future performance optimizations and contributed to more robust machine learning workflows within the project.