Home › Evidence › Records › cfg-0142
⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.

cfg-0142

backfilled
citable URL: https://halobench.com/records/cfg-0142/ — this address never moves; the anchor /records/#cfg-0142 keeps resolving
modelQwen3.6-35B-A3B-MTP-UD-Q4_K_M.gguf · UD-Q4_K_M · rev 0b21525e972670ed59e1812e170b27c26355381f0656ecc4e25617ece7dac58b
engineggml-org/llama.cpp 7077abbe14c510cb829c93a1328c2815b5805ebd · ROCm0/gfx1151 · host aihydra (igpu)
flagsHU-004 long-prompt guard n2 target arm on raw llama-server: model /home/aihydra/models/qwen3.6-35b-a3b-mtp/Qwen3.6-35B-A3B-UD-Q4_K_M.gguf, -ngl 999 -dev ROCm0 -fa on -ctk f16 -ctv f16 -c 204800 --load-mode none -b 2048 -ub 512 --parallel 1 --host 127.0.0.1 --port 8114 --spec-type draft-mtp --spec-draft-n-max 2 -ctkd f16 -ctvd f16. Server binary build-rocm/bin/llama-server, build 50 commit 7077abb — the same binary/path used by HO-009 cfg-0134 with native model MTP/NextN. No ngram/external drafter, no proxy/llama-swap, no n_max>=4.
templatenot recorded at test time
treeupstream — stock
aged evidence — reconstructed from the archive. HU-004 reviewer-admitted long-prompt guard target arm only. Guard-cleared evidence is a guard-pass result, not a capability, throughput, or production result. No n2 d262144 substitution and no n_max>=4 safety claim are made.

Capability basis

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

Runs on this config (1)

rundatekindsuitemetrics
run-05522026-08-21guardguard-hu004-longprompt-8kto40ktool_call_success=1 · empty_arg_calls=0 · tasks_passed=10 · tasks_total=10

Cited by — computed at build time, never stored

claims clm-0101