CUBE-AI 8.0 INTERNAL ERROR: list index out of range

[Alpha_Lee]

STM32 CubeAI v1.7.0 (STM.ai v8.0.0-19389)
error: list index out of range

• This is a model created and transformed using pytorch

  

  • Is the model too big?

[Issamos]

Hello @Alpha_Lee 

Can you rewrite your question in English for two reasons. First, English is the official language of the community. Also, our community members and experts can easily understand your question and answer it.

Best regards.

II

[Alpha_Lee]

Yeah, you're right.I rewrote it.

[fauvarque.daniel]

Please share your model (could be with fake weights) so we can reproduce the issue.

The error you see is very likely a limitation in the current version of the tool, are you using 1D layers ?

Regards

[Alpha_Lee]

The model size is 20M and I can't upload it. Is it because the model is too big?

[fauvarque.daniel]

try to zip it, alternatively you can share a link on this page

Thanks

 

[Mai1]

Have you tried quantizing the model?

[fauvarque.daniel]

Issue reproduced on our current development branch, bug opened

[fauvarque.daniel]

mbg82, your test.onnx model is malformed, 

/SA/MatMul_1_output_0_DequantizeLinear has input [1x8], axis is 0, and x_scale is [8].

axis should be 1 (not 0)

Also onnxruntime gives an error

Regards

Daniel