AutoProbe

AutoProbe — run artifacts This deposit contains the raw run artifacts produced by **AutoProbe**, an agentic pipeline in which a large-language-model agent automatically designs, implements, and iteratively improves quantitative evaluation **probes** for machine-learning training pipelines. Given a project's `train.py`, the agent proposes a probe, integrates it, and rewrites the training code round by round until the probe metric crosses a target, reverting any round that improves the probe at the expense of the model's own utility.