Qwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)

Qwen3.8-27B runs on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw) with 24K context and both K/V on int8, while the delta net stays FP16 and decoding peaks at 70 tok/s.