comparing models without fooling yourself for Phi

[nanobrook]

Thanks, that helps. The pattern around MoE serving has been pretty similar for me. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.