cfg-0139
backfilled
citable URL: https://halobench.com/records/cfg-0139/ — this address never moves; the anchor /records/#cfg-0139 keeps resolving
| model | GLM-4.7-Flash-Q4_K_M.gguf · Q4_K_M · rev 29837ed2c0fc5f51981adf8ac8083fcf80743c598381f13e9f06cbad0498b174 |
| engine | ggml-org/llama.cpp 3653e6d · ROCm0/Strix-Halo · host aihydra (igpu) |
| flags | HG-004 Phase-1 boundary diagnostic fingerprint: raw /home/aihydra/src/llama.cpp/build/bin/llama-server on 127.0.0.1:8104, llama.cpp checkout ce7689f875f724ece08b1a9fbb2a83971120b4d2 with clean tree, server binary sha256 17f55ef427a56ebd98aabf0d3c48e43a93e384e1db5075c5c7bed5ed6a0aad01, version 1 (3653e6d), GLM-4.7-Flash Q4_K_M model sha256 29837ed2c0fc5f51981adf8ac8083fcf80743c598381f13e9f06cbad0498b174 bytes 18312339808, --load-mode none, --no-mmap, -ngl 999, -dev ROCm0, -fa on, -ctk f16, -ctv f16, -c 32768, -n 4096, --parallel 1, --jinja, --reasoning-format auto, pinned Tau2/native-template task-1 path with frozen and embedded template sha256 d63ad536c3c81880043e22ec7fd08db42b4d8fb7c89c7138bc562bfa25281375, task-1 sha256 365091613cf731aaadac059d7a39b32b95a070bf358b4bd9f04ad836eec1a614, tools-14 sha256 bdba5499d11340d02e92688d62f942f2f702667368d04e52772520aa8fa7ea22, simulator openrouter/anthropic/claude-haiku-4.5 attempts=1. No tau2 Phase-2, benchmark, throughput, or capability result is admitted from this diagnostic. |
| template | d63ad536c3c81880043e22ec7fd08db42b4d8fb7c89c7138bc562bfa25281375 |
| tree | upstream — stock |
aged evidence — reconstructed from the archive. Reviewer-admitted HG-004 Phase-1 boundary diagnostic only from /home/aihydra/bench-results/hg004-glm47-phase1-boundary-r1/hg004-glm47-phase1-boundary-r1-20260820T231235Z and /Users/wardenmac/.openclaw/workspace/halobench/reconcile/HG-004-phase1-boundary-review.md. The record preserves one raw llama-server HTTP 200 response capture whose normalized assistant content was empty with no tool_calls, non-empty reasoning_content, and finish_reason length. Tau2 Phase-2 was not launched, so no score, task pass count, throughput, capability verdict, or production recommendation is represented.
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-0544 | 2026-08-20 | guard | hg004-phase1-boundary-diagnostic | runner_invalid=true |