Mixtral aka the 8x7B the "sparse mixture of experts" one is not the same as, eg. Mistral-7B which is still very, very good, just not quite hitting the mark on some things.
I still couldn't run Mixtral 8x7B on an M1 Macbook Pro with 32Gb ram, so maybe I am indeed doing it wrong? Or are there better quantized versions available now or..?
