llama.cpp Metal on Apple M1 Ultra (64-core GPU, 128 GB): llama 7B Q4_0 tg128 (discussion #4167, c1d0e7a)

COMMUNITY llama-bench posted to llama.cpp discussion #4167. llama-bench -p 512 -n 128 -ngl 99, mean of the reporter's repeats, single stream, batch size 1. The number is per-request, NOT an aggregate across concurrent requests. Model is the canonical llama-2-7b GGUF (6.74B) the thread standardises on. Posted by @marcingomulkiewicz, comment 18191340 (2026-08-28), in reply to the updated-instructions request: M1 Ultra, 16+4 CPU, 64-core GPU, 128 GB. Backend MTL,BLAS, 16 threads. Build c1d0e7a00 (b10621). The 8e672ef rows already on main for this chip are a different binary; the reporter filed this specifically because the old numbers are not comparable, and ggerganov noted the chip was already scored on the old build.

Same machine and harness as the other rows in this comment. Recorded mean only; the reporter's +/- is in the methodology. format_id is null for F16 because data/quantization has no F16 GGUF record.