I’m running a fully coherent model fully on my IPhone 15 pro max! LiquidAI LFM-2.6B in Q8 It did a web search and tool call and dropped facts. 22 tok/s prefill 2-3 tok/s decode But I could have used q4 for more speed