ocean-SR-training

Automate ocean super-resolution model training across multiple GPUs by invoking ocean_sr_train_start with dataset_root, log_dir, model_name, and scale. The workflow coordinates model selection, dataset handling, and training orchestration across single or multi-GPU clusters.

Supported architectures for ocean super-resolution workflows include FNO2d, SwinIR, UNet2d, and SRNO. These models integrate into the training pipeline with safety checks, checkpoints, and break-glass recovery workflows.

Ocean super-resolution training handles out-of-memory errors through robust OOM protection and memory management safeguards. The workflow includes safe restart capabilities and break-glass recovery to resume training without losing progress.

You can run full-dataset inference and generate visualizations after training by using ocean_sr_train_visualize and ocean_sr_train_report. The predict mode allows ramping back to training after evaluating the generated plots.

Monitor training status for ocean super-resolution models using the ocean_sr_train_status tool, which provides explicit parameter confirmation and progress monitoring. This ensures continuous tracking throughout the training, evaluation, and reporting phases.