Worked on enhancing language model dataset handling and input processing in the apple/axlearn repository, focusing on improving the reliability and scalability of data ingestion pipelines. Leveraged Python and data processing techniques to update the grain library version and implement targeted edge-case tests, ensuring robust handling of diverse input scenarios. Applied fixes associated with the grain upgrade to stabilize the dataset pipeline, reducing the likelihood of failures during language model training. Automated testing was introduced to prevent regressions and maintain data integrity. The work laid a foundation for future scalability, emphasizing careful testing and incremental improvements in machine learning data workflows.