john-rocky/coreai-model-zoo

Downloadable models and conversion recipes for Apple's Core AI on iPhone and Mac. Chat, vision, speech and generative models with per-model validation records, Swift examples through CoreAIKit, and a downloadable Mac app.

What it solves

It provides a collection of pre-converted AI models and the exact recipes needed to run them on Apple's Core AI framework (iOS 27 / macOS 27). This allows developers to easily integrate high-performance, offline AI capabilities—such as LLMs, vision-language models, and speech-to-text—into Apple apps without having to manually handle the complex conversion process from original model formats.

How it works

The project maintains a "zoo" of .aimodel bundles and corresponding recipe.toml files. These recipes define the exact configuration used for conversion, allowing anyone to rebuild the model bundle using zoo_convert.py. To ensure quality, the project uses a gating system (coreai_gate.py) that compares the output of the converted model against a reference fp32 model to prove numerical parity. Integration is simplified via the CoreAIKit library, which enables downloading and running these models with a single line of Swift code.

Who it’s for

Apple platform developers (iOS and macOS) who want to deploy state-of-the-art open-source AI models on-device while ensuring they are optimized for Apple hardware.

Highlights

  • Diverse Model Support: Includes text LLMs (Qwen, Gemma, LFM), Vision-Language Models (VLM), speech-to-text (Whisper), and specialized models like safety classifiers and robotics (BitVLA).
  • Verified Parity: Uses a rigorous gating process to ensure converted models produce the same tokens as the original source models.
  • Advanced Optimizations: Features custom Metal kernels for 1.58-bit ternary LLMs and mixed-bit quantization for high-speed inference on iPhone GPUs.
  • One-Line Integration: Works with CoreAIKit to handle model downloading, caching, and execution with minimal code.

Related

  • Project
    john-rocky/CoreML-Models

    A comprehensive collection of pre-converted Core ML models and sample apps for implementing on-device AI on iOS and macOS.

  • Dispatch
    Swift Transformers Release: Run On-Device LLMs on Apple Devices

    Hugging Face released the alpha swift-transformers Swift package, swift-chat demo app, updated exporters and transformers-to-coreml tools, and Core ML versions of Llama 2 7B and Falcon 7B, enabling developers to run LLMs on-device in Apple apps.

  • Project
    apple/coreai-models

    A toolkit for exporting, authoring, and running open-source AI models on Apple silicon using the Core AI framework for macOS and iOS.

  • Project
    soniqo/speech-swift

    Speech Swift is an open‑source Swift package that provides dozens of on‑device speech AI models (ASR, TTS, translation, voice agents, enhancement, etc.) for Apple Silicon. It runs via Core ML and the MLX framework, supports hundreds of languages, offers multiple quantised sizes, and can be added to a project with a single Swift‑Package‑Manager dependency. The repo includes extensive docs, pre‑trained models, and example code for transcription, streaming dictation, and UI components, all without any cloud calls.

  • Project
    ailia-ai/ailia-models

    A collection of 419 pre-trained AI models across various domains like vision, audio, and text, all runnable via a unified CLI with automatic weight downloads.