Thomas Keller

Specifically, the issue arises with the `scale` factor being calculated to zero, then dividing by scale in the course of `IntQuant.to_int()` (shown below): ``` class IntQuant(brevitas.jit.ScriptModule): ... def __init__( self,...

> But I did not face the same issue if I use `Int8Bias` are you facing the same issue with it? Yes, Int8Bias gives the same results for me.

+1

Bumping this, would really appreciate a simple example beyond the basic functionality validation of the `test_merge_bn.py` file.

> However, from a quick glance, it seems that it removes the "brevitas exception". Which might break FINN code? The PR adds a "domain aliasing" system ([Lines 36-39 in registry.py](https://github.com/fastmachinelearning/qonnx/pull/204/files#diff-c27dc0c878cb0a1449b37eda6da99a09a6d2f2238e1371ae7dad07d7579ab692R36-R39))...