How to get a Qoperator format ONNX model after quantization?

[JiliangNi]

How to get a Qoperator format ONNX model after quantization?

[e-said]

Hi @JiliangNi

I'm not aware of Qoperator support in AIMET. However, you can obtain QDQ format in your ONNX using use_embedded_encodings=true with AIMET's ONNX export feature. If you're unfamiliar with QDQ format, you can find more information in this link

PS: Please note that AIMET QDQ format is supported only for int8 quantization (W8A8) due to limitation in the ONNX opset version related to the Torch version (1.13) of AIMET

[JiliangNi]

So AIMET does not support Qoperator format in ONNX, right?

Is there any method which could convert QDQ to Qoperator in ONNX?

[quic-mangal]

@JiliangNi, we don't have the support to convert currently.