Worked on stabilizing dataset loading for MATH leaderboard evaluations by addressing configuration issues in the red-hat-data-services/lm-evaluation-harness and swiss-ai/lm-evaluation-harness repositories. Focused on resolving dataset_path resolution problems, two targeted bug fixes were delivered to ensure the evaluation harness could reliably locate and load the MATH dataset. Leveraging configuration management skills and expertise in YAML, updated task templates and configuration files to point to the correct dataset repositories. These changes improved evaluation reliability and reduced manual setup for leaderboard runs, ensuring consistent access to the MATH dataset and streamlining the evaluation process across both environments without introducing new features.