Community
Anemll Forge 0.2.1 runs Qwen3.8-27B on Apple M6 Neural Engine with 192K context
Version 0.2.1 adds INT8 K/V cache and a smaller drafter, cutting 0.86 GB of ANE memory. On a 32GB M6 Mac Mini, the author cites about 27 tok/s decode and 21.4 GB total memory at 96K context, with speed varying by content type.
Every outlet that ran it 1 article
r/LocalLLaMACommunityLead
M6 Mac Mini owners (32GB): now you can run Qwen3.8-27B in high quality on the Apple Neural Engine