liveBook · Manning

Core ML

As explained in Quantization and Fast Inference

Core ML is Apple's framework for integrating machine-learning models into applications across its platforms.

Core ML is Apple's universal machine-learning runtime for deploying models across macOS, iOS, iPadOS, and visionOS. Its .mlpackage format uses flatbuffers and supports vision, audio, and the prefill stage of large language models. section 11.3

Deployment hinges on compute-unit selection:

  • Provide an MLComputeUnits hint; it sets a ceiling, while Core ML chooses per-operation fallbacks according to the Apple Neural Engine's fixed operator-support table. section 11.3
  • Distinguish Core ML from MLX: Core ML serves broad Apple-platform deployment, while MLX targets autoregressive decoding on M-series hardware. section 11.3

How Quantization and Fast Inference covers it

How deep it goes
Explains package compatibility and compute-unit fallback behavior without detailing the operator-support table.
Examples
Names vision, audio, and large-language-model prefill as supported workload examples.
What you can do after
Operate deployment by setting an MLComputeUnits ceiling and accounting for per-operation fallback.
How it connects
Contrasts Core ML with MLX and ties runtime fallback to the Apple Neural Engine.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Quantization and Fast Inference. The section links point to the parts of the book it draws on.