cfg-0162
backfilledfork
citable URL: https://halobench.com/records/cfg-0162/ — this address never moves; the anchor /records/#cfg-0162 keeps resolving
| model | Qwen3.6-35B-A3B-MTP-UD-Q4_K_M.gguf · UD-Q4_K_M · rev 0b21525e972670ed59e1812e170b27c26355381f0656ecc4e25617ece7dac58b |
| engine | ggml-org/llama.cpp 2586f6eddae19bef3dd21a1a0b109cca7bcf1c32 · Vulkan0/Strix-Halo (strix-halo-v0610 fork via RADV, NOT ggml-org ROCm0) · host aihydra (igpu) |
| flags | HO-013 cell2-plain-rea-ON sentinel fingerprint: PLAIN (NO MTP spec flags) with reasoning ON (-rea on) on the v0.6.10 fork/Vulkan runtime 2586f6edd. sentinel-rea fingerprint records base_flags: -ngl 999 -fa 1 -c 32768 --load-mode none --parallel 1 --jinja -rea on -n 4096 --slots -ctk f16 -ctv f16 -b 2048 -ub 512 -t 16, reasoning=on, spec=plain (NO spec_flags). This is the reasoning-content sentinel control arm. |
| template | not recorded at test time |
| tree | fork — carrying con-0002, con-0009 |
aged evidence — reconstructed from the archive. DISPOSITION-ONLY. This plain-rea-ON control FAILED the reasoning-content sentinel (G6 FAIL, rc=2): assistant_content_len=0 (EMPTY) despite reasoning_content_len=1846, num_tool_calls=2, finish_reason=tool_calls. The reasoning-ON path is BLOCKED on this fork/stack — no tau2 was run on any reasoning-ON arm, and the whole cell2/cell3 reasoning set was stopped/cancelled (see clm-0113). No capability/score. Scope strict to single model/quant/build.
Capability basis
unmeasured — performance numbers on this config stand on a guard alone, not an established capability
Runs on this config (1)
| run | date | kind | suite | metrics |
|---|---|---|---|---|
| run-0600 | 2026-08-23 | guard | reasoning-sentinel-c32768 | runner_invalid=true |
Cited by — computed at build time, never stored
claims clm-0113