video1.0<iframe src="https://www.loom.com/embed/42d4627b476c4282a53e636a940ad6d9" frameborder="0" width="1114" height="835" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>8351114Loomhttps://www.loom.com8351114https://cdn.loom.com/sessions/thumbnails/42d4627b476c4282a53e636a940ad6d9-384049df892d1de4.gif255.837This Loom demonstrates an evaluation workflow for a real GCP Enclave demo, showing how mock data and private assets are handled. The speaker restarts an in-memory notebook run where folders act like actors, removes prior files for a fresh execution, and then sets up the model owner and benchmark owner roles. They use a model named “Quen 0.5 billion,” with mock dataset and helper code that researchers can browse to develop against the same data structure, while private weights and private inference code remain inside the enclave. The flow includes sharing mock and benchmark data to the enclave, which can be slow due to copying, and notes that researchers have access to mock datasets and samples but not private weights.