World Wires · story 52964 · unverified · 1 source(s)

Qwen3.8-27B running on iPhone 18 Neural Engine at 15-70 tok/s

A demonstration shows the Qwen3.8-27B model running on an iPhone 18 Neural Engine at 15 to 70 tokens per second using 1.5 bpw quantization.

Open in the desk

Coverage

What this site indexes