[Crash] Crash while loading AlibabaNLP/gte-base ONNX model

[vivekg-ai]

Describe the issue

We are seeing a crash report while loading certain models (e.g. AlibabaNLP/gte-base - : https://huggingface.co/Alibaba-NLP/gte-base-en-v1.5)

The crash is seen on macOS, and here is the crash location -

To reproduce

Try loading the given model from the URL - https://huggingface.co/Alibaba-NLP/gte-base-en-v1.5

See the crash while inferencing.

Urgency

No response

Platform

Mac

OS Version

14.5

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

https://github.com/thebrowsercompany/onnxruntime-swift-package-manager

ONNX Runtime API

Obj-C

Architecture

ARM64

Execution Provider

Default CPU

Execution Provider Library Version

No response

[yufenglee]

@vivekg-ai, could you please share a call stack and how is the model converted to ONNX?

[vivekg-ai]

We are using the ONNX model directly available here - https://huggingface.co/Alibaba-NLP/gte-base-en-v1.5/tree/main/onnx

Unfortunately, the call stack doesn't have much since we are using the release build of the official SPM package of the ONNX Runtime from here - https://github.com/thebrowsercompany/onnxruntime-swift-package-manager

[tianleiwu]

That only indicates that the model is failed to create session.

@vivekg-ai, There are multiple models in https://huggingface.co/Alibaba-NLP/gte-base-en-v1.5/tree/main/onnx. Which one did you use?

[vivekg-ai]

We tried -

[tianleiwu]

I tried it in my windows machine. It is good to create session from python API

import onnxruntime
ort_session = onnxruntime.InferenceSession("model.onnx", providers=['CPUExecutionProvider'])

Could be related to Mac. @skottmckay, @edgchen1 to take a look at onnxruntime-swift-package-manager.