lenet.cpp: FC layers differ from lenet.py, output mismatches getting_started.md — is a known issue?

Hi, quick question: in lenet/lenet.cpp, the three FC layers use Dims2{400,120} with MatrixOperation::kNONE, while lenet.py uses reshape(120,400) + TRANSPOSE. My -d output doesn't match the getting_started.md reference (e.g. 0.0949172... vs 0.0949623..., off by ~0.005-0.008). Is this a known issue, or already fixed on the latest master? I can provide a full repro and a fix patch. My clone is at commit 89db562.