gradient-check

This skill diagnoses vanishing or exploding gradients in a hybrid quantum-classical training path by tracing gradient flow from loss through parameters, comparing magnitudes, and proposing fixes like residual connections or gradient scaling.

AMP boundary instability during quantum model training is caused by gradient flow disruptions at the mixed precision boundary. Diagnostics map these paths and propose gradient scaling adjustments to stabilize the training run.

Diagnose V6 collapse in trainable quantum models by checking scripts like model.py and enhanced_training.py to map gradient paths, flag unstable gradients, and surface actionable fixes to prevent optimization stalls.

Yes, you can compare quantum and classical gradient magnitudes to prevent training stalls. Running a diagnostic pass surfaces bottlenecks in gradient flow and reports concrete stabilization actions like initialization changes.

Residual connections for gradient stabilization in hybrid quantum training are limited by underlying architectural constraints in scripts like trainable_quantum_model.py. They cannot fix fundamental V6 collapse if initialization or scaling is also misconfigured.