pick a model

Conifer measured your machine. The tokens-per-second figures are from a calibration kernel that ran here, not from a published baseline.

preliminary estimate — calibrates on first generation