QuantLinear cannot compute F.Linear beacuse it is flattening the input

[balditommaso]

Describe the bug

The new version of Brevitas may have a bug when you apply intger quantization to a dense layer, I have tryed to quantise different models from torchvision and when you reach the last linear layer the multiplication fail.

Reproducibility

• Can be reproduced consistently.
• [ x] Difficult to reproduce.
• Unable to reproduce.

Screenshots/Code Example/Log Output
RuntimeError: mat1 and mat2 shapes cannot be multiplied (131072x1 and 512x1000)

please complete the following information:

• Brevitas version: last
• PyTorch version: 2.5
• Operating System / platform: Linux

[Giuseppe5]

Thanks for sharing.
Without a script it is difficult for us to reproduce. Does the problem disappear with the previous release of Brevitas?

[balditommaso]

Solved, it was my issue