Worked on stabilizing training workflows in the openvinotoolkit/training_extensions repository by addressing configuration issues in MaskDINO and MonoDETR projects. Focused on configuration management using YAML, the developer removed obsolete parameters from AdaptiveTrainScheduling and EarlyStoppingWithWarmup, reducing misconfigurations and runtime errors during model training. The changes were validated in the development branch and documented in the project’s changelog, improving traceability for downstream users. This work enhanced the reliability and reproducibility of training pipelines, streamlining experiment setup for future development. The approach demonstrated attention to detail in configuration hygiene and contributed to more maintainable and robust model training workflows across projects.