Worked on enhancing cross-device shared memory management in the pytorch/pytorch repository, focusing on unified handling for XPU, HPU, and CUDA tensors. Used C++ and Python to replace per-backend allowlists with a streamlined check, simplifying non-CPU memory sharing logic and correcting nested CUDA tensor semantics. Expanded and updated unit tests to ensure correctness of is_shared and share_memory behavior across devices. In the intel/torch-xpu-ops repository, stabilized CI tests by adapting to recent PyTorch API changes, maintaining test reliability. These efforts reduced runtime errors, improved multi-device throughput, and strengthened alignment with upstream changes, supporting more robust backend development and testing.