Home › Evidence › Records › cfg-0151
⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.

cfg-0151

citable URL: https://halobench.com/records/cfg-0151/ — this address never moves; the anchor /records/#cfg-0151 keeps resolving
modelQwen3.5-122B-A10B-UD-Q4_K_M-00001-of-00003.gguf · UD-Q4_K_M
engineggml-org/llama.cpp ce7689f · ROCm0/gfx1151 · host aihydra (igpu)
flagsHO-012 production-optimisation matrix Cell-1 baseline (b2048/ub512, q8_0/q8_0 KV) + Cell-2 depth probe fingerprint. llama-server build ce7689f (llama.cpp-kvfix tree), -ngl 999 -fa 1 -c 32768 --parallel 1 --load-mode none --jinja -rea off -ctk q8_0 -ctv q8_0 -b 2048 -ub 512 -t 16 -n 4096, ROCm0 Radeon 8060S gfx1151 122880 MiB. Cell-2 depth via llama-bench -p 16384 -n 256 -d 262144 -r 1 on the same b2048/ub512 q8_0/q8_0 signature. From per-execution command files (server-c1-b2048-ub512.command.txt and cell2-depth/c2-d262144.command.txt).
templatenot recorded at test time
treeupstream — stock

Capability basis

unmeasured — performance numbers on this config stand on a guard alone, not an established capability

Runs on this config (3)

rundatekindsuitemetrics
run-05792026-08-23performanceho012-122b-prod-matrix-r2.cell1decode_tps=21.8436 · ttft_ms=1420.66
run-05852026-08-23performanceho012-122b-prod-matrix-r2.cell2-depthdecode_tps=8.447554 · prefill_tps=62.2502
run-05882026-08-23guardguard-c32768-depth8000tasks_passed=4 · tasks_total=4

Cited by — computed at build time, never stored

claims clm-0111