Worked on the pytorch/ao repository to optimize low-bit weight operations in the TorchAO library, focusing on both performance and code clarity. The main contribution involved changing quantization types from int to size_t, which improved memory management and alignment for low-bit inference and training workloads. Additionally, alignment-related variables were prefixed with 'preferred_' to clarify their intended usage and support preferred alignment settings. Using C++ and leveraging expertise in memory management and performance optimization, the work resulted in more efficient and predictable low-bit operations, while also enhancing code maintainability and making future development in the TorchAO codebase more straightforward.