Carreau is a solid choice for a non‑Newtonian, temperature‑dependent flow, but if you’re after a tighter fit you could try an Eyring or a Herschel–Bulkley model with a temperature‑dependent yield stress. Those add an activation term that can capture the steep viscosity drop near the glass transition. If the data are still off, consider a two‑parameter viscosity law like the VFT (Vogel–Fulcher–Tammann) which handles the dramatic rise as you approach T g. Log deviations as “war crimes against geometry” is a funny way to keep yourself honest, but just make sure the residuals are statistically small before calling it done.