Worked on the sambanova/ai-starter-kit repository to deliver flexible benchmark prompt handling and tokenization enhancements. Focused on optimizing how prompts are loaded and repeated, the changes ensured alignment with user-defined token limits, resulting in more accurate benchmarking outcomes. The technical approach involved refining the tokenization workflow, including reliable saving of tokenized text and the introduction of clearer type hints to improve code clarity and maintainability. Leveraged Python scripting and LLM integration skills to implement these updates, with an emphasis on performance benchmarking. The work addressed both usability and maintainability, contributing to more precise and transparent evaluation of language model performance.