cfg-0149
backfilledfork
citable URL: https://halobench.com/records/cfg-0149/ — this address never moves; the anchor /records/#cfg-0149 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-009-v0610 PLAIN (reasoning-OFF) capability fingerprint on the v0.6.10 fork/Vulkan runtime, mirroring cfg-0144 flags exactly (f16/f16 KV, -ngl 999 -fa on -c 32768 --load-mode none --parallel 1 --jinja -b 2048 -ub 512 -t 16, no spec/MTP flags) PLUS the tau2-protocol agent settings and -rea off on the tau2 path. temp 0, seed 42, max_tokens 4096. Same model/quant/flags as cfg-0144 but built on the v0.6.10 fork under Vulkan/RADV instead of upstream 7077abb ROCm0/gfx1151 — repo + backend + build all changed together, so this is a stack effect, not a build-only change. |
| template | not recorded at test time |
| tree | fork — carrying con-0002, con-0009 |
aged evidence — reconstructed from the archive. Recorded from the raw tau2 artifacts and the v0.6.10 fork MANIFEST (payload source 2586f6edd, Vulkan/RADV). The -rea off delta is the deliberate DIAG-recommended fix already applied in cfg-0144/0145. Scoped to the v0.6.10-fork-on-Vulkan vs 7077abb-upstream-on-ROCm0 capability recovery question only; no DFlash2/stock-MTP/n_max>=4/reasoning-on/alt-model/ production-throughput claim.
Capability basis
measured on this config
Runs on this config (2)
| run | date | kind | suite | metrics |
|---|---|---|---|---|
| run-0575 | 2026-08-23 | guard | guard-c32768-depth8000 | tasks_passed=4 · tasks_total=4 |
| run-0576 | 2026-08-23 | capability | tau2-bench-airline@668d3bc | tasks_passed=22 · tasks_total=26 · mean_score=0.8461538461538461 |
Cited by — computed at build time, never stored
claims clm-0110