cfg-0138
backfilledfork
citable URL: https://halobench.com/records/cfg-0138/ — this address never moves; the anchor /records/#cfg-0138 keeps resolving
| model | Qwen3.8-27B-Q8_0.gguf + Qwen3.8-27B-DFlash2-Q8_0.gguf · Q8_0 + DFlash2-Q8_0 · rev 57484e3196aaff8dbbd71c666158b00206baa2bc134e4f16ae90fc8cbdfbeea2 + 7f1c9a31a6ed40044c69f6508b50fd63b87abd8e1fb7fe4290303df549153751 |
| engine | ggml-org/llama.cpp 0b0f35d0ed745bd6e40eb248e320cacbb1a546ab · Vulkan0/Strix-Halo · host aihydra (igpu) |
| flags | HO-004 DFlash2 Q8_0 width-4 Stage A fingerprint: same target/runtime/backend/KV/batch/load-mode as cfg-0136 plus external drafter /home/aihydra/models/qwen38-27b-dflash2/Qwen3.8-27B-DFlash2-Q8_0.gguf with -md DFlash2-Q8_0 -ngld 999 --spec-type draft-dflash --spec-draft-n-max 4. No stock MTP, no ngram/proxy/llama-swap/ROCm, no alternate target quant, no KV quant substitution. |
| template | not recorded at test time |
| tree | fork — carrying con-0002, con-0003 |
aged evidence — reconstructed from the archive. Stage A activation/safety arm only. Positive DFlash2 draft acceptance counters were reviewer-admitted, but the arm failed the Stage A sentinel gate on code/toolish prompts; no guard, Stage B performance, energy-efficiency, or production recommendation claim is admitted.
Capability basis
measured on this config
Runs on this config (1)
| run | date | kind | suite | metrics |
|---|---|---|---|---|
| run-0543 | 2026-08-20 | guard | stageA-c32768-varied-prompts | runner_invalid=true |