Developed robustness and reporting enhancements for the DLC Runner in the thunlp/SIR-Bench repository, focusing on improving the reliability of distributed machine learning workflows. Leveraged Python to refine the loading of model configurations and optimize the handling of DLC job commands, ensuring smoother execution across diverse datasets. Enhanced pre-training summarization and expanded metrics reporting provided richer observability and clearer evaluation of dataset performance. These updates improved reproducibility and facilitated faster issue diagnosis, supporting better stakeholder visibility into benchmark results. The work demonstrated strong skills in data analysis, distributed systems, and model training, addressing key challenges in large-scale machine learning infrastructure.