Error in running tutorial/hello_world

[ngdymx]

Hi team,

I came to the step 3 Quantize Model and encountered an issue in the following figure.
Could you help me? thanks a lot!

[vgodsoe]

Hi @ngdymx
Can you tell me a bit more about your setup that resulted in this error? I'm trying to reproduce it on my end but I don't get the same message?

Here is the quantization settings that are working for me:

vai_q_onnx.quantize_static(
    input_model_path,
    output_model_path,
    calibration_data_reader=None,
    quant_format=vai_q_onnx.QuantFormat.QDQ,
    calibrate_method=vai_q_onnx.PowerOfTwoMethod.MinMSE,
    activation_type=vai_q_onnx.QuantType.QUInt8,
    weight_type=vai_q_onnx.QuantType.QInt8,
    enable_ipu_cnn=True,
    extra_options={'ActivationSymmetric': True}
)

Thanks,
Victoria

[ngdymx]

Hi,

Thanks.
I just follow the tutorial to setup the environment.
Um, maybe because I'm using a windows system, some paths are not set up very well, it is my guess.

I tried to change the calibrate_method to other mode, it works, wierd.

[vgodsoe]

@ngdymx Just to confirm, the updates are working for you now?

[ngdymx]

Under the vai_q_onnx.PowerOfTwoMethod.MinMSE:, it still doesn't work.
It works under another mode like vai_q_onnx.PowerOfTwoMethod.NonOverflow:.