Worked on the EvolvingLMMs-Lab/lmms-eval repository to deliver a new evaluation feature that improves experiment flexibility and reproducibility. Developed and integrated a command line --offset option using Python scripting and argparse, allowing users to start dataset evaluation from any specified index. This addition enables targeted benchmarking and more efficient data processing workflows without disrupting existing evaluation logic, ensuring backward compatibility and a low-risk deployment. The work focused on enhancing the pipeline’s usability for reproducible research, with careful attention to seamless integration and minimal risk. No major bugs were addressed during this period, as the focus remained on feature delivery.