CoreML API Reference

Comprehensive API reference for CoreML model integration, MLTensor operations, coremltools conversion, and state management.

Overview

This reference covers the complete CoreML API surface for on-device machine learning:

• MLModel - Model loading, prediction, lifecycle
• MLTensor - Pipeline stitching between models (iOS 18+)
• coremltools - Python package for conversion and compression
• State management - KV-cache for LLM inference
• MLComputeDevice - Runtime compute availability

When to Use This Reference

Use when you need:

• ☑ API details for MLModel, MLTensor, or coremltools
• ☑ State management patterns for LLMs
• ☑ Compute device availability checking
• ☑ Model conversion and compression APIs
• ☑ Performance profiling APIs

Example Prompts

• "What are the MLModelConfiguration options?"
• "How do I use MLTensor for post-processing?"
• "What's the API for checking Neural Engine availability?"
• "How do I create an MLComputePlan?"
• "What compression APIs does coremltools provide?"

Documentation Scope

This reference documents the coreml-ref skill. Use it when you need precise API signatures and configuration options.

• For implementation patterns and decision trees, see CoreML skill
• For troubleshooting model issues, see CoreML Diagnostics

Key APIs

MLModel Loading

// Async load (preferred)
let model = try await MLModel.load(contentsOf: url)

// With configuration
let config = MLModelConfiguration()
config.computeUnits = .all
let model = try await MLModel.load(contentsOf: url, configuration: config)

MLTensor Operations (iOS 18+)

let tensor = MLTensor([[1.0, 2.0], [3.0, 4.0]])
let result = (tensor * 2.0).softmax()
let array = await result.shapedArray(of: Float.self)

State for KV-Cache

let state = model.makeState()
let output = try await model.prediction(from: input, using: state)

Compute Availability

let hasNeuralEngine = MLModel.availableComputeDevices.contains {
    if case .neuralEngine = $0 { return true }
    return false
}

Related

• CoreML Skill — decision trees and patterns for common workflows
• CoreML Diagnostics — troubleshooting when things go wrong
• Foundation Models — Apple's built-in LLM when you don't need custom models

Apple Documentation

• Core ML
• Core ML Tools