issue with ApplyAlpha for conv2d layers

[m-glowacki]

I'm attempting a hls synthesis of a simple quantised CNN model, featuring:

• conv_2d layers
• dense layers
• batch normalisation

the model is quantised during training, with no alpha scaling (alpha=1)

the synthesis breaks with this exception:

  File "/software/ys20884/miniconda3/envs/ml_env/lib/python3.8/site-packages/hls4ml/model/layers.py", line 105, in _validate_attributes
    raise Exception('Unexpected value of attribute "{}" of layer "{}" ({}). Expected {}, got {} ({})'
Exception: Unexpected value of attribute "n_in" of layer "q_conv2d_alpha" (ApplyAlpha). Expected <class 'int'>, got <class 'NoneType'> (None)

this is the layer in question:

Layer name: q_conv2d, layer type: QConv2D, input shapes: [[None, 20, 12, 1]], output shape: [None, 18, 10, 4]

I'm on the current main branch

I previously had no issues with the same model on release 0.6.0

thanks for taking a look.

[bo3z]

Hi @m-glowacki. Thanks for raising this issue. I believe this has been addressed in PR #612 . Could you check out the branch and see if it resolves the problem?

[m-glowacki]

thanks @bo3z! Indeed this has fixed things, issue closed.