sop-ft-orchestrate

Automate SOP monitoring pipeline fine-tuning by orchestrating end-to-end workflows across data augmentation, DDM temporal segmentation, Cosmos-Reason VLM training, evaluation, and root-cause remediation. Configure YAML inputs with dataset paths and target accuracy thresholds to start autonomous iterative model improvement.

Yes, this orchestration skill supports independent DDM-Net and VLM retry axes. It applies configuration fixes from root-cause analysis findings and uses watchdogs to detect training failures, hangs, timeouts, and resource issues for autonomous recovery.

Before running an SOP model evaluation workflow, you need annotated training and evaluation video datasets, configured training and evaluation services, GPU resources, YAML configuration files, and Python dependencies including PyYAML and asyncpg.

Iterative root-cause remediation works by analyzing evaluation outputs and RCA reports to apply configuration fixes automatically. The pipeline maintains run state and iteration budgets, continuing corrective iterations until target accuracy is met or the budget is exhausted.

Manage reproducible run states for DDM temporal segmentation pipelines by maintaining configuration snapshots, phase status, evaluation outputs, and progress files. The orchestration skill uses watchdog execution and safe code overrides to ensure reproducible iterative model improvement.

Your SOP fine-tuning pipeline stops when the configured iteration budget is exhausted. The orchestration skill iteratively applies data augmentation and root-cause remediation until the desired accuracy thresholds are met or the maximum allowed iterations are reached.