cfg-0145
backfilled
citable URL: https://halobench.com/records/cfg-0145/ — this address never moves; the anchor /records/#cfg-0145 keeps resolving
| model | Qwen3.6-35B-A3B-MTP-UD-Q4_K_M.gguf · UD-Q4_K_M · rev 0b21525e972670ed59e1812e170b27c26355381f0656ecc4e25617ece7dac58b |
| engine | ggml-org/llama.cpp 7077abbe14c510cb829c93a1328c2815b5805ebd · ROCm0/gfx1151 · host aihydra (igpu) |
| flags | HO-009-AB n2/MTP (reasoning-OFF) capability fingerprint: same model/runtime/ backend as cfg-0134 plus -rea off on the tau2 path. MTP engaged via --spec-type draft-mtp --spec-draft-n-max 2 -ctkd f16 -ctvd f16 (draft KV stays f16 = no KV quant, matching clm-0097/cfg-0134; --spec-draft-n-max 2 alone is a silent no-op on build 7077abb, per hborchestrator fingerprint resolution). f16/f16 target KV, -ngl 999 -fa on -c 32768 --load-mode none --parallel 1 --jinja -b 2048 -ub 512 -t 16. temp 0, seed 42, max_tokens 4096. n_max=2 ONLY; no n_max>=4, no KV quant, no alt backend. |
| template | not recorded at test time |
| tree | upstream — stock |
aged evidence — reconstructed from the archive. The -rea off delta is the deliberate DIAG-recommended fix, not a deviation. Scoped to the plain-vs-n2 capability comparison on build 7077abb only. No n2 quality-equivalence, production recommendation, or throughput claim beyond the plain-vs-n2 capability comparison on this card.
Capability basis
measured on this config