In May 2025, Ankit Gulati enhanced the pytorch/pytorch repository by developing cross-device compatibility testing for the FSDP2 module. He refactored existing test cases to be device-agnostic, enabling validation across CUDA, HPU, and XPU accelerators. This work improved the reliability of distributed training features by ensuring consistent behavior on multiple hardware platforms. Using Python, PyTorch, and distributed computing techniques, Ankit focused on test engineering and CI automation to streamline multi-accelerator validation. The depth of the refactoring reduced device-specific issues and accelerated feature readiness, contributing to more robust and maintainable unit testing for PyTorch’s distributed training stack.