Not if it's 5-10x slower than a remote inference server. Mac prefill latency is exhausting.
Oh tell me more about prefill latency.