During November 2025, Dakhare developed a configurable model training validation and checkpointing system for the NVIDIA/physicsnemo repository. This work focused on enhancing training workflows by introducing periodic validation on a separate dataset, with automatic checkpoint saving to improve model assessment and reliability. Using Python and PyTorch, Dakhare integrated the new validation workflow into the existing training script, allowing users to adjust validation frequency and checkpointing parameters for greater reproducibility and traceability. The implementation established a foundation for improved metrics tracking and data-driven model selection, addressing the need for robust evaluation and consistent results in machine learning model development.