Describe the issue

I set onnxruntime to verbose to understand why the CoreML provider is that slow. Then I figured out it does convert the ONNX to mlmodel every time without a cache.

2024-08-15 23:01:33.924908 [I:onnxruntime:, model_builder.cc:926 SaveModel] Writing CoreML Model to /var/folders/nz/t7blntb55gv6k4hzbd91kl5h0000gn/T/onnxruntime-72B02264-8413-46DF-8067-45E9808043DF-16638-000199F2EF486AAE.model.mlmodel

To reproduce

Just use set_default_logger_severity(0)

I use a ONNX with opset 17 - but the issue is not related to a single model

Urgency

No response

Platform

Mac

OS Version

14.5

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.18.0

ONNX Runtime API

Python

Architecture

ARM64

Execution Provider

CoreML

Execution Provider Library Version

No response