Worked on the pytorch/pytorch repository to deliver cross-device compatibility testing for the FSDP2 module. Refactored existing test cases to be device-agnostic, enabling validation across CUDA, HPU, and XPU accelerators. This approach improved the reliability of distributed training tests by ensuring consistent behavior regardless of hardware, and enhanced the speed and confidence of feature validation in continuous integration workflows. Leveraged Python, PyTorch, and distributed computing principles to expand test coverage and reduce device-specific issues. The work focused on robust unit testing and CI automation, contributing to faster readiness of new features and more dependable multi-accelerator support in PyTorch.