Yuanxiulong Yxl developed advanced data processing and training efficiency features for the modelscope/ms-swift and alibaba/ChatLearn repositories using Python and deep learning techniques. For ms-swift, Yuanxiulong introduced dynamic bucketing for persistent cache padding and a flattened data collator, optimizing memory usage and reducing training overhead by minimizing unnecessary padding. In ChatLearn, Yuanxiulong engineered data loading optimizations for distributed systems, including sorting samples within global batches and implementing a skip-generation mode to accelerate reproducibility. These solutions improved throughput, training stability, and cost efficiency, demonstrating strong skills in cache management, distributed systems, and machine learning engineering over a focused two-month period.