Home › Evidence › Records › cfg-0025

cfg-0025

backfilled
citable URL: https://halobench.com/records/cfg-0025/ — this address never moves; the anchor /records/#cfg-0025 keeps resolving
modelQwen3.5-122B-A10B-UD-Q4_K_M-00001-of-00003.gguf · UD-Q4_K_M
engineggml-org/llama.cpp 3653e6d6d547ec763317d9ecd0ace334a7e21359 · rocm · host aihydra (igpu)
flags-ngl 999 -fa on -c 32768 --parallel 1 --load-mode none --jinja -rea off -ctk f16 -ctv f16
templatenot recorded at test time
treeupstream — stock
aged evidence — reconstructed from the archive. tau2 serving config for the Aug 10-11 protocol arms (0.545 headline, anchor pair, sim-sensitivity, task9-regression). Flags come from the runner scripts — powered.sh on-box for the Aug 9-10 arms, bench/queues/queue-2026-08-11.sh / queue-f / queue-g in this repo — with the build asserted fatal at launch, not trusted. Memory block, HF revision and chat-template hash were not recorded at test time; GGUF sha256 is ledgered in audits/environment.lock-aihydra-2026-08-12.json.

Capability basis

measured on this config

Runs on this config (5)

rundatekindsuitemetrics
run-01002026-08-09capabilitytau2-bench-airline@668d3bctasks_passed=12 · tasks_total=22 · mean_score=0.5455
run-01052026-08-11capabilitytau2-bench-airline@668d3bctasks_passed=9 · tasks_total=9 · mean_score=1
run-01092026-08-11capabilitytau2-bench-airline@668d3bctasks_passed=9 · tasks_total=9 · mean_score=1
run-01102026-08-11capabilitytau2-bench-airline@668d3bctasks_passed=7 · tasks_total=9 · mean_score=0.7778
run-01112026-08-11capabilitytau2-bench-airline@668d3bctasks_passed=1 · tasks_total=8 · mean_score=0.125

Cited by — computed at build time, never stored

model pages qwen35-122b