One thing you can also do which is really cool:
you can set up pipeline parallelism between that mac and your primary machine over ethernet, and partially load a larger dense model (sadly, only denses) like qwen27B across your multiple machines for linear layer-by-layer processing
