[Bug] fp8 quantization, weight_error is empty, but qdq_err is normal

[zccyman]

Describe:

        weight_type=QuantType.QFLOAT8E4M3FN,
        activation_type=QuantType.QFLOAT8E4M3FN,
        calibrate_method=CalibrationMethod.Distribution,
        extra_options=dict(
            QDQKeepRemovableActivations=True,
            ActivationSymmetric=False,
            WeightSymmetric=False,
            AddQDQPairToWeight=True,
            QuantizeBias=True,
            ForceQuantizeNoInputCheck=True,
            TensorQuantOverrides=dict(),
        ),

weight errors is empty, but qdq_err is normal.

Comparing weights of float model vs qdq model.....
weight errors: 

------------------------------------------------

Augmenting models to save intermediate activations......
------------------------------------------------

Running the augmented floating point model to collect activations......
------------------------------------------------

Running the augmented qdq model to collect activations......
2024-06-20 06:53:28.508970448 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer 'conv1.bias_quantized_zero_point'. It is not used by any node and should be removed from the model.
2024-06-20 06:53:28.508988421 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/conv1/Conv_output_0_scale'. It is not used by any node and should be removed from the model.
2024-06-20 06:53:28.508992868 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/conv1/Conv_output_0_zero_point'. It is not used by any node and should be removed from the model.
2024-06-20 06:53:28.508997639 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer 'conv1.bias_quantized_scale'. It is not used by any node and should be removed from the model.
------------------------------------------------

Comparing activations of float model vs qdq model......
qdq_err: 

output 0.0839491831621686
input 0.036262167281220226
xmodel_err: 

output 0.15343184633604956
input 0.036262167281220226

[github-actions]

This issue has been automatically marked as stale due to inactivity and will be closed in 30 days if no further activity occurs. If further support is needed, please provide an update and/or more details.