JevMade

Sign in
← Back to experiments

SDKs & clients

CoreAI Model Zoo

John-rocky’s model zoo brings open decision models to Apple’s Core AI runtime, with tests of their probabilities and device timing.

Source screenshot of CoreAI Model Zoo
SOURCE SCREENSHOTFull screenshot ↗

What it does

The zoo includes ports of Mapika decider, Kev, Liquid d1, OpenThai, Laya and a Qwen scorer. These are existing models packaged by John-rocky, not newly trained models. Supported devices differ between ports. A result measured on a Mac tells you nothing conclusive about performance on an iPhone.

How you can use it

The maker provides examples for particular devices and checks answers against the original models. Use those to choose a port for an Apple app that needs local choices, ratings or yes/no answers instead of written replies. Some ports accept images. Check which hardware and Apple app-building software your chosen model needs.

Maker-reported (not independently measured by JevMade): john-rocky reports: john-rocky reports 37.8 ms per Kev-0.8B question on iPhone 18 Pro. · john-rocky reports: john-rocky reports 34.3 ms per d1-3B question on M4 Max. · john-rocky reports: john-rocky reports matching Laya answer choices on 81 checks; iPhone timing is not yet measured.

Primitives
choice, score, noul
Platform
Apple Core AI
Added
Project created

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.