alt Hacker NewsWhile what can be done with less bits per parameter may be interesting, this is not really the important number for an LLM.
Much more important is the total amount of memory needed by an LLM, which can be decreased either by reducing the number of bits per parameter or the number of parameters. Arithmetic circuits are already cheap so for the inference cost the amount of memory transfers is more important.
It is not clear yet which number of bits per parameter allows the minimization of the total memory requirement, at a given LLM quality.
I thought the multiplications during training (backpropagation) are quite computationally expensive. The method from the paper above doesn't need any multiplications.