Lookup "occupancy chart antidepressant", study the chart. There is an exponential drop-off at lower doses. Hence to taper off without feeling it, one needs to cut smaller and smaller doses, (or liquify) which may be impractical.
Gosh, instead of wasting time vibing another slop project, stop using Claude Code, use pi or whatever, own the harness, and explore the plethora of models out there. You guys are software engineers, yes?
Yes it's between this and Gemma 4 31B which is much slower, but looks like it won't ever get an upgrade. I have to conclude that the MoE variants are unreliable, and MTP sometimes just can't get tricky formatting right.