Implementation:Microsoft Onnxruntime CoreMLFlags

Knowledge Sources	Description
Source File	java/src/main/java/ai/onnxruntime/providers/CoreMLFlags.java
Repository	Microsoft/onnxruntime

Domains

• Machine Learning Runtime
• Execution Provider Configuration
• Apple CoreML

Overview

CoreMLFlags is a public enum providing bitwise flags for configuring the Apple CoreML execution provider. It implements the OrtFlags interface, allowing flags to be aggregated into a single integer bitmask via EnumSet. These flags control CPU-only mode, subgraph enablement, Apple Neural Engine requirements, static shape constraints, MLProgram creation, and GPU usage.

Description

The enum defines six flags:

• CPU_ONLY (1): Use only CPU, disabling GPU and Apple Neural Engine. Not recommended for production.
• ENABLE_ON_SUBGRAPH (2): Enables CoreML on subgraphs.
• ONLY_ENABLE_DEVICE_WITH_ANE (4): Only enable CoreML if the device has an Apple Neural Engine.
• ONLY_ALLOW_STATIC_INPUT_SHAPES (8): Restrict CoreML to inputs with static shapes (dynamic shapes may reduce performance).
• CREATE_MLPROGRAM (16): Create an MLProgram instead of a NeuralNetwork model. Requires Core ML 5+.
• CPU_AND_GPU (32): Exclude the Apple Neural Engine, use CPU and GPU only.

Code Reference

Source Location

// File: java/src/main/java/ai/onnxruntime/providers/CoreMLFlags.java
// Package: ai.onnxruntime.providers

Signature

public enum CoreMLFlags implements OrtFlags {
    CPU_ONLY(1), ENABLE_ON_SUBGRAPH(2), ONLY_ENABLE_DEVICE_WITH_ANE(4),
    ONLY_ALLOW_STATIC_INPUT_SHAPES(8), CREATE_MLPROGRAM(16), CPU_AND_GPU(32);

    public final int value;
    public int getValue();
}

Import

import ai.onnxruntime.providers.CoreMLFlags;

I/O Contract

Inputs

Name	Type	Description
(enum constant)	CoreMLFlags	The flag to include in the configuration

Outputs

Name	Type	Description
getValue()	int	The native bitmask value for this flag

Usage Examples

import ai.onnxruntime.*;
import ai.onnxruntime.providers.CoreMLFlags;
import java.util.EnumSet;

try (OrtSession.SessionOptions opts = new OrtSession.SessionOptions()) {
    // Add CoreML with MLProgram and static shapes only
    opts.addCoreML(EnumSet.of(
        CoreMLFlags.CREATE_MLPROGRAM,
        CoreMLFlags.ONLY_ALLOW_STATIC_INPUT_SHAPES));

    OrtEnvironment env = OrtEnvironment.getEnvironment();
    try (OrtSession session = env.createSession("/path/to/model.onnx", opts)) {
        // Model will use CoreML MLProgram backend
    }
}

Related Pages

• OrtSession.java - SessionOptions.addCoreML() accepts CoreMLFlags
• OrtProvider.java - CORE_ML provider constant
• NNAPIFlags.java - Similar flags enum for Android NNAPI