Worked on the llm-d-benchmark repository to enhance resource management in model serving by introducing a new configuration option for the vllm command. Developed the NUM_CPU_BLOCKS setting, which allows users to tune CPU memory usage, resulting in more predictable memory footprints during both benchmarking and live serving scenarios. This feature was implemented using Bash and Shell scripting, leveraging skills in configuration management and DevOps practices. The work focused on providing greater control over system resources, addressing the need for efficient and reliable model deployment. No bug fixes were recorded during this period, with efforts concentrated on this targeted feature improvement.