pycoreml 0.2.0

CoreML inference from Python — powered by Swift & ApplePy

PyCoreML

CoreML inference from Python — powered by Swift & ApplePy.

Direct access to Apple's CoreML framework with explicit compute unit selection (CPU, GPU, Neural Engine).

macOS only — requires Swift 6.0+ toolchain

Install

pip install -e .

Usage

import pycoreml

# Load and describe a model
info = pycoreml.load_model("/path/to/model.mlmodelc")
desc = pycoreml.model_description("/path/to/model.mlmodelc")

# Run prediction
result = pycoreml.predict("/path/to/model.mlmodelc", {"x": 42.0})

# Predict with explicit compute unit (the killer feature)
result = pycoreml.predict_with_options(
    "/path/to/model.mlmodelc",
    {"x": 42.0},
    "cpuAndNeuralEngine"  # or "cpuOnly", "cpuAndGPU", "all"
)

API

Function	Returns	Description
load_model(path)	str	Load model and return description
model_description(path)	dict[str, str]	Model metadata
predict(path, inputs)	dict[str, float]	Run inference
list_compute_units()	list[str]	Available compute units
predict_with_options(path, inputs, unit)	dict[str, float]	Predict with compute unit control

Examples

See pycoreml/examples/demo.py for a full demo.

License

BSD-3-Clause © Jagtesh Chadha — see LICENSE.