Data Science & Analytics

TAO Run DEFT Pas

The skill automates iterative refinement of NVIDIA TAO CLIP or SigLIP image‑text retrieval models on attribute‑labelled data, preparing datasets, evaluating zero‑shot performance, analyzing attribute gaps, mining caption‑space k‑NN pairs, selecting history‑aware samples, retraining, and re‑evaluating until a KPI, validation plateau, or iteration budget is met. It provides data scientists and product teams with a repeatable, KPI‑driven improvement loop that requires no knowledge of DEFT or People Attribute Search terminology.

Publisher / source

NVIDIA Skills (Verified)

Upstream record

NVIDIA/skills

Version

Version 3

Published

2026-09-30

Publication review

Approved for publication

Content fingerprint

a596e8a7542a31d2b8c219434226f45eb5e62156fd6f1dd0494ef8288885d54a