01
Model discovery:
Focuses on PyTorch, Core AI.
AI & Machine Learning / 23:44
Integrate on-device AI models into your app using Core AI
AI & Machine Learningiosipadosmacoswatchos
Key Points
- Covers a curated collection of popular open-source models - including Qwen, Mistral, SAM3, and more - optimized for Apple silicon using the new Core AI Framework.
- Explains how to download, run, and benchmark models on the Mac, and integrate them into an app with just a few lines of code.
- Covers a new workflow for model compilation and on-device specialization to speed up first-time model load.
- Find out how to profile and optimize runtime performance with Core AI tools in Xcode.
- The session moves through App concept: camera-based vocab learning, Model discovery, Getting models with the Core AI models repository, Integration, Writing the Swift integration code, Diagnosing model specialization latency, and related wrap-up guidance.
- Key concepts include Core AI, SAM3, PyTorch, CoreAIImageSegmenter, CoreAILM, CoreAISegmentation, Foundation Models, Instruments.
Condensed Flow
02
Getting models with the Core AI models repository:
Focuses on Core AI.
03
Writing the Swift integration code:
Focuses on CoreAIImageSegmenter.
04
Diagnosing model specialization latency:
Focuses on Core AI.
05
Deployment:
Focuses on Core AI.
06
Ahead-of-time (AOT) compilation:
Focuses on Core AI.
More Detail
Additional details
- Detailed flow: App concept: camera-based vocab learning -> Model discovery -> Getting models with the Core AI models repository -> Integration -> Writing the Swift integration code -> Diagnosing model specialization latency -> Deployment -> Ahead-of-time (AOT) compilation -> ...
- APIs and concepts to recognize: Core AI, SAM3, PyTorch, CoreAIImageSegmenter, CoreAILM, CoreAISegmentation, Foundation Models, Instruments, RAW, CoreAILanguageModel, HuggingFace, GitHub.
- Version and support notes focus on iOS 27.0, macOS 27.0.
- Implementation focus: PyTorch, Core AI, CoreAILM, CoreAISegmentation, CoreAIImageSegmenter, Foundation Models, Instruments, RAW, CoreAILanguageModel, HuggingFace.