Adding an extra Custom Attribute to an existing operator

[adityat77]

Describe the issue

I want to latch an extra attribute to each node of a simple MNIST ONNX model.
I'm able to add custom attributes and save the model. However, while loading the model for Inference, I'm getting this:
**Error Unrecognized attribute: CustomAttr for operator Conv**

I've attached a link for the model for reference.

https://drive.google.com/file/d/1votf5NdwU2YHamhCsyrEDi3nkSCBPBjZ/view?usp=sharing

To reproduce

Load the model using this:

onnxruntime.InferenceSession("model.onnx")

Urgency

No response

Platform

Linux

OS Version

4.18.0

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

1.16.1

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response

[adityat77]

is there a better way to do this / any suggestions are appreciated

[hariharans29]

The short answer is this can't be done as it violates the ONNX spec/interface for the operators found in the MNIST model. The spec/interface for each operator may be found here. The error you see while loading the mutated model is exactly complaining about this - it doesn't recognize the new attribute as it isn't supported by the spec. Is there any reason why you need to add a custom attribute to every node in the model ?

[adityat77]

i'm trying to write an e2e application with onnxruntime ir.
will explore alternatives, thanks.

[pranavsharma]

What do you plan to do with the custom attribute? What's your use case?

[hariharans29]

Please re-open the issue with more details