Worked on optimizing data handling for uint5 formats within the pytorch/ao repository, focusing on improving the efficiency of packing and unpacking operations. Applied advanced bit manipulation and low-level programming techniques in C++ to reduce the computational overhead associated with uint5 data processing. The new implementation streamlined critical data paths, lowering CPU cycles and increasing throughput for workflows reliant on uint5 values. This work laid the foundation for broader support of uint5 formats and future performance enhancements. The changes targeted performance optimization, emphasizing efficient bitwise operations and careful resource management to accelerate execution in applications utilizing custom data widths.