Worked on the deepinv/deepinv repository to enhance GPU-enabled workflow reliability by refactoring the get_freer_gpu utility. The update introduced support for both PyTorch API and nvidia-smi methods, allowing users to select their preferred GPU detection approach and optionally suppress warnings. This change addressed environment adaptability and reduced fragility in GPU selection across diverse deployment scenarios. Comprehensive tests were added to validate the new arguments and behaviors, ensuring robust system integration. Documentation was updated to clarify usage and warn users about potential issues. The work leveraged Python and focused on GPU management, testing, and clear documentation to improve CI stability and user guidance.