Home › Evidence › Records › cfg-0185
⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.

cfg-0185

fork
citable URL: https://halobench.com/records/cfg-0185/ — this address never moves; the anchor /records/#cfg-0185 keeps resolving
modelQwen3.8-27B-IQ4_XS-ALL-IMATRIX-Q8-OUT-MTP.gguf · IQ4_XS-ALL-IMATRIX-Q8-OUT · rev ilintar/qwen3.8-27b-gguf-strix-halo (file 16,110,851,680 bytes)
enginepwilkin/llama.cpp d3b5cc43d1fcfce891f2de94d5274ee40eceb21c · rocm · host aihydra (igpu)
flags-dev ROCm0 -ngl 999 -fa on -fit off --load-mode none -c 49152 -b 2048 -ub 512 --parallel 4 --cont-batching --jinja --reasoning-effort low --spec-type draft-dflash --spec-draft-model Qwen3.8-27B-DFlash2-IQ4_XS.gguf --spec-draft-device ROCm0 --spec-draft-ngl 99 --spec-draft-n-min 0 --spec-draft-n-max 3 --spec-draft-p-min 0.10 --metrics --port 5803 (runtime env: ENABLE_RETAINED_PM4=1 -> DEBUG_HIP_GRAPH_PM4=1, HSA_OVERRIDE_GFX_VERSION=11.5.1, GGML_HIP_ENABLE_UNIFIED_MEMORY=1; custom ROCr+HIP runtime via LD_LIBRARY_PATH)
samplingtemperature 0 · max_tokens 8192
templatenot recorded at test time
treefork — carrying con-0018, con-0019

Capability basis

measured on this config

Runs on this config (3)

rundatekindsuitemetrics
run-06652026-09-06capabilitytau2-bench-airline@668d3bcempty_arg_calls=0 · tasks_passed=20 · tasks_total=24 · mean_score=0.8333
run-06662026-09-06guardempty-output-monitor@dflash-pm4tool_call_success=1 · turns_before_degradation=none — 0 EMPTY over 16 probes across 66 min (09:27-10:33Z) · empty_arg_calls=0
run-06672026-09-06performanceslotpin-mslot@dflash-pm4decode_tps=47.3