cfg-0146
backfilledfork
citable URL: https://halobench.com/records/cfg-0146/ — this address never moves; the anchor /records/#cfg-0146 keeps resolving
| model | Qwen3.8-27B-Q8_0.gguf · Q8_0 · rev 57484e3196aaff8dbbd71c666158b00206baa2bc134e4f16ae90fc8cbdfbeea2 |
| engine | ggml-org/llama.cpp 2586f6eddae19bef3dd21a1a0b109cca7bcf1c32 · Vulkan0/Strix-Halo · host aihydra (igpu) |
| flags | HO-004-DIAG v0.6.10 plain/control Stage-A sentinel fingerprint (re-tests the v0.5.x base-platform empty-content defect after the v0.6.10 MTP-rollback and DSpark re-land): /home/aihydra/src/strix-halo-v0610/vulkan/llama-server, MANIFEST.txt "Built 2026-08-22T10:47:53Z source 2586f6eddae19bef3dd21a1a0b109cca7bcf1c32", target Qwen3.8-27B Q8_0 (sha256 57484e...), --load-mode none, -ngl 999, -fa on, -b 2048, -ub 512, f16/f16 KV, --parallel 1, --jinja, -c 32768, --temp 0, --seed 4242, no speculation, no MTP, no DFlash2 drafter, no ngram/proxy/llama-swap/ROCm. Same portable Vulkan family and KV/batch/load-mode as cfg-0136 but on the v0.6.10 build. |
| template | not recorded at test time |
| tree | fork — carrying con-0002, con-0009 |
aged evidence — reconstructed from the archive. Recorded from the runner command line and MANIFEST.txt at review time; the full build-commit/Mesa/source lineage is in the artifact MANIFEST (mesa d18d598e2). Sentinal outcome measured: plain/control PASS Stage-A (empty-assistant 0, empty-tool-call 0, single-token-EOS 0, code-valid 3/3, tool-correct 3/3) plus the 4/4 house guard, so the null-update diagnostic resolved. This config is ONLY a Stage-A sentinel/gate record: it carries no throughput, no Stage-B production-depth performance, and no capability score.
Capability basis
measured on this config
Runs on this config (1)
| run | date | kind | suite | metrics |
|---|---|---|---|---|
| run-0572 | 2026-08-23 | guard | stageA-c32768-varied-prompts | tasks_passed=4 · tasks_total=4 |
Cited by — computed at build time, never stored
claims clm-0109