qualcomm/ai-hub-models

onboard

Onboard a new model to Qualcomm AI Hub Models. Guides authoring a standalone folder (model.py, app.py, demo.py, test.py, manifest.yaml) that the qai-hub-models CLI can install, export, and evaluate. Ships `supported_precisions: [float]`; quantization is a…

Idioma del texto original: inglés

add-quantization

Add a quantized precision to a float recipe that already passes `validate-on-device`. Tries w8a8, falls back to w8a16, or gives up. Hard-requires a dataset + evaluator (authors them under supervision if missing). Use after `validate-on-device`.

Idioma del texto original: inglés

onboard-internal

Onboard a new model to Qualcomm AI Hub Models as an in-tree recipe (lives under src/qai_hub_models/models/<id>/, appears in the public catalog, has a scorecard entry). This is a thin delta on top of the base `onboard` skill — it inherits all authoring…

Idioma del texto original: inglés

validate-on-device

Compile, profile, and (if evaluator wired) evaluate a float recipe on a Snapdragon device via AI Hub. Iterates on failures using `.claude/docs/on-device-debugging.md`. Use after `onboard` produces a `validate`-green float recipe, and before `add-quantization`.

Idioma del texto original: inglés