train

Hybrid-model training orchestration coordinates baseline and enhanced trainable-quantum paths using checkpoint handling and platform-aware execution. It manages legacy runs, ablations, and result summarization automatically.

You can run hybrid-model training on macOS CPU or CUDA environments. The orchestration performs platform checks to validate your setup and selects the appropriate execution path, ensuring compatibility across local machines and Colab.

To launch a baseline training run with checkpoints, execute python -m src.train with default settings. Console output will display progress while checkpoint validation ensures your training state is preserved and recoverable.

Baseline training paths use legacy execution via python -m src.train, while enhanced trainable-quantum paths run through train_v7.py. Both are orchestrated with checkpoint handling, but the enhanced path integrates quantum model training capabilities.

Running local ablations for hybrid models requires executing train_ablation_local.py. The orchestration skill manages this process alongside baseline and enhanced paths, capturing metrics and runtime behavior to reproduce conclusions.

Hybrid-model training workflows need checkpoint validation to ensure training states are recoverable and gradient health is monitored. This observability captures metrics and flags drift, helping reproduce conclusions across different platform executions.