GitHub - john-rocky/coreai-model-zoo: Community model zoo + knowledge base for Apple Core AI (iOS/macOS 27): Qwen3.5 & Gemma 4 converted end-to-end, verified on-device (iPhone 17 Pro GPU/ANE), conversion gotchas, custom Metal kernels, Swift runner · GitHub
/" data-turbo-transient="true" />
Skip to content
Search or jump to...
Search code, repositories, users, issues, pull requests...
-->
Search
Clear
Search syntax tips
Provide feedback
--><br>We read every piece of feedback, and take your input very seriously.
Include my email address so I can be contacted
Cancel
Submit feedback
Saved searches
Use saved searches to filter your results more quickly
-->
Name
Query
To see all available qualifiers, see our documentation.
Cancel
Create saved search
Sign in
/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up
Appearance settings
Resetting focus
You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.
Dismiss alert
{{ message }}
john-rocky
coreai-model-zoo
Public
Notifications<br>You must be signed in to change notification settings
Fork
Star<br>47
main
BranchesTags
Go to file
CodeOpen more actions menu
Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit
History<br>66 Commits<br>66 Commits
apps
apps
conversion
conversion
knowledge
knowledge
official
official
swift
swift
zoo
zoo
.gitignore
.gitignore
LICENSE
LICENSE
README.md
README.md
View all files
Repository files navigation
CoreAI-Model-Zoo
LLMs converted to Apple Core AI (.aimodel, iOS 27 / macOS 27) — downloadable, verified<br>on-device, with the conversion code and a knowledge base. Successor to<br>CoreML-Models.
Models
Model<br>Download (.aimodel)<br>License
Qwen3.5-0.8B<br>🤗 qwen3.5-0.8B-CoreAI<br>Apache-2.0
Qwen3.5-2B<br>🤗 qwen3.5-2B-CoreAI<br>Apache-2.0
Qwen3.6-35B-A3B (MoE, Mac-only)<br>🤗 Qwen3.6-35B-A3B-CoreAI<br>Apache-2.0
Gemma 4 E2B (text, incl. official-QAT int4)<br>🤗 gemma-4-E2B-CoreAI<br>Gemma
Gemma 4 E4B (text, official-QAT int4)<br>🤗 gemma-4-E4B-CoreAI<br>Gemma
LFM2.5-1.2B-Instruct<br>🤗 LFM2.5-1.2B-CoreAI<br>LFM Open License v1.0
Granite 4.0-H 1B / 350M<br>🤗 granite-4.0-h-CoreAI<br>Apache-2.0
Qwen3-VL (vision-language)<br>🤗 2B · 4B · 8B<br>Apache-2.0
Gemma 4 E2B vision (VL) (image+text)<br>vl/ in 🤗 gemma-4-E2B-CoreAI<br>Gemma
RF-DETR nano/small/medium/large (object detection, no NMS)<br>🤗 RF-DETR-CoreAI<br>Apache-2.0
Decode throughput (tok/s, greedy; output top-1 exact vs the Hugging Face reference)
iPhone 17 Pro · GPU<br>iPhone 17 Pro · ANE<br>M4 Max · GPU
Qwen3.5-0.8B<br>71.9<br>14.7<br>210
Qwen3.5-2B<br>29<br>161
LFM2.5-1.2B<br>45.4<br>276.5
Granite 4.0-H 1B<br>36.3<br>136.5
Gemma 4 E2B<br>30.3 (QAT 30.7)<br>77.0 (QAT 78.9)
Gemma 4 E4B (official QAT)<br>15.1<br>55.8
Gemma 4 E2B VL (image+text, official QAT)<br>25.5<br>82.4
Qwen3.6-35B-A3B (MoE, 35B/~3B active, Mac-only)<br>30.9
Measured on the iOS 27 / macOS 27 beta, Apple's coreai-pipelined GPU engine, zero custom<br>kernels (ANE column excepted). Prefill, sizes, per-model caveats: zoo/.
Qwen3.6-35B-A3B (MoE, 35B/~3B active) — 30.9 tok/s is expert-gather-bound in the<br>current beta; zoo/qwen3.6.md
RF-DETR — 33–39 FPS live on iPhone 17 Pro, 8.6–19.1 ms/frame on M4 Max;<br>zoo/rf-detr.md
Gemma 4 E2B VL — same text decoder + a 3-line image splice;<br>zoo/gemma4-vl.md
CoreAIChat (apps/) — the zoo's models running on-device on iPhone.
Repository layout
Dir<br>What
zoo/<br>Model cards — configurations, sizes, parity, measured throughput.
knowledge/<br>Verified notes on the framework: conversion, compression, stateful KV, custom Metal kernels, AOT, compute-unit rules, the Swift runtime.
conversion/<br>Re-authored models + convert / verify / compress scripts (PyTorch → .aimodel).
swift/<br>CoreAIRunner — a Swift package that drives .aimodel LLM bundles, including architectures beyond the standard runtime.
apps/<br>SwiftUI on-device chat apps (iOS 27): CoreAIChat (Gemma 4 E2B GPU/ANE/⚡ + Qwen3.5 / Qwen3.5-2B / LFM2.5 / Granite ⚡pipelined, one picker) + QwenChatFast (Qwen3.5 static kernels) with in-app model download.
Start here
Run a model on device → knowledge/swift-runtime.md + the model card
Convert a model → knowledge/conversion-guide.md
Compress → knowledge/compression.md
Make it fast → knowledge/custom-metal-kernels.md · knowledge/performance-ceiling.md
Known beta issue (in-graph KV-write crash; workarounds + the input-mask escape) → knowledge/coreai-beta-mpsgraph-kvwrite-bug.md — FB23024751 / apple/coreai-models#5
License
BSD-3-Clause (LICENSE). Re-authored model code derives from Apple's BSD-3-Clause<br>coreai_models and retains its notices. Model weights follow their own licenses (see each<br>Hugging Face repo).
About
Community model zoo + knowledge base for Apple Core AI (iOS/macOS 27): Qwen3.5 & Gemma 4 converted end-to-end, verified on-device (iPhone 17 Pro GPU/ANE), conversion gotchas, custom Metal kernels, Swift...