A weekend with TensorFold on a MacBook: the engine mattered, the quant did not

A 27B dense model on my MacBook decodes at about 26 tokens a second. That is fine for chat and...