MIXTRAL 8X22B ON A $400 CPU FASTER THAN READING SPEED (GITHUB DISCUSSION)

Summary from the TLDR AI newsletter

Thanks to recent advances from Mozilla's Llamafile project, it is now possible to run inference for the flagship model from Mistral at 20 tokens per second on a commodity CPU.