---

name: autovp_training_loop description: Run a reduced AutoVP prompt optimization step while proving the source classifier remains frozen.

AutoVP Training Loop

Use this skill to recover or test AutoVP's parameter-efficient adaptation mechanism. It is appropriate when a frozen classifier callable, prompt module, label mapper, and small dataset are available.

Inputs

Training examples, labels, prompt parameters, frozen classifier weights/callable, label mapping, learning rate, and step count.

Outputs

Loss before/after, prompt parameters before/after, classifier frozen check, predictions, and accuracy.

Workflow

Apply visual prompt, run frozen classifier, map logits to targets, compute loss, update only prompt or mapping parameters, and record trace evidence. Never mutate source classifier weights.

Validation

Run included tiny deterministic optimizer test.