Contributed to the liguodongiot/transformers repository by developing quantized tensor parallelism support, enabling scalable training with reduced-precision data types. Leveraged Python and deep learning frameworks to optimize memory and compute efficiency, conditionally setting gradient requirements based on data type to streamline training throughput. Addressed maintainability by simplifying state_dict processing, removing DTensor type checks to focus on torch.Tensor types, which reduced edge-case risks and improved code clarity. The work enhanced the reliability of model loading in PyTorch-backed workflows and positioned the repository for broader adoption of quantization techniques, demonstrating depth in machine learning and parallel computing within a focused two-month period.