Worked on the vllm-project/llm-compressor repository to deliver a feature focused on improving hardware portability for deep learning workloads. Migrated all PyTorch CUDA calls to the unified torch.accelerator API, enabling the codebase to support multiple hardware accelerators such as CUDA and ROCm through a single interface. This refactor streamlined the main code path, reducing hardware-specific dependencies and enhancing compatibility across diverse environments. The work was implemented using Python and leveraged expertise in CUDA, PyTorch, and machine learning. No bugs were addressed during this period, with efforts concentrated on the feature migration and ensuring robust cross-hardware support for future development.