Worked on the NVIDIA/Megatron-LM repository to enhance benchmarking capabilities for multimodal AI models by adding support for the OCRBenchV2 and RD-TableBench datasets. Developed Python evaluation scripts and integrated them into the existing evaluation framework, enabling more comprehensive and automated performance assessments. Focused on data processing and scripting to ensure seamless compatibility with the current benchmarking infrastructure. The work expanded the range of datasets available for model evaluation, allowing for broader and more nuanced analysis of model performance. Utilized skills in benchmarking, data processing, and Python to deliver a robust solution that supports ongoing research and development efforts.