Home › Evidence › Records › cfg-0136
⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.

cfg-0136

backfilledfork
citable URL: https://halobench.com/records/cfg-0136/ — this address never moves; the anchor /records/#cfg-0136 keeps resolving
modelQwen3.8-27B-Q8_0.gguf · Q8_0 · rev 57484e3196aaff8dbbd71c666158b00206baa2bc134e4f16ae90fc8cbdfbeea2
engineggml-org/llama.cpp 0b0f35d0ed745bd6e40eb248e320cacbb1a546ab · Vulkan0/Strix-Halo · host aihydra (igpu)
flagsHO-004 plain/control Stage A fingerprint: /home/aihydra/src/strix-halo-v065-run-20260820195059, asset e9e8d0067604b02ff3d69946ede758a4708d6b75791aa55f8340c0ae43c5a234, manifest 034b652092d91f54a55f6fb7bc0101a2669da3864fd5d732a562ca8fc2826f11, Mesa d18d598e2, Qwen3.8-27B Q8_0 target, --load-mode none, -ngl 999, flash attention on, -b 2048, -ub 512, f16/f16 KV, --parallel 1, temp 0, seed 4242, no speculation, no MTP, no DFlash2 drafter, no ngram/proxy/llama-swap/ROCm.
templatenot recorded at test time
treefork — carrying con-0002
aged evidence — reconstructed from the archive. Reviewer-admitted HO-004 negative Stage A record from /home/aihydra/bench-results/ho004-qwen38-dflash2-matched-r1. The arm loaded/responded but failed the Stage A sentinel gate on code/toolish prompts; guard and Stage B performance were not run, and no throughput or production claim is admitted.

Capability basis

measured on this config

Runs on this config (1)

rundatekindsuitemetrics
run-05412026-08-20guardstageA-c32768-varied-promptsrunner_invalid=true