A suite of OLMo-style models from the paper "Disentangling Geometry, Performance, and Training in Language Models"