During December 2025, TJ Dals delivered the MFU Metrics Framework for Transformer Performance Monitoring in the jeejeelee/vllm repository. This framework was developed using Python and focused on enabling precise estimation of Model FLOPs Utilization (MFU) by calculating FLOPs and memory IO, providing detailed insights into resource usage for large-scale transformer workloads. TJ Dals integrated the framework with existing observability configurations, supporting end-to-end performance tracking and data-driven optimization. The work demonstrated depth in data analysis and performance optimization, laying a foundation for proactive tuning and capacity planning. Unit testing ensured reliability, and cross-team collaboration enhanced the framework’s integration.