Home › Evidence › Records › cfg-0187
⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.

cfg-0187

fork
citable URL: https://halobench.com/records/cfg-0187/ — this address never moves; the anchor /records/#cfg-0187 keeps resolving
modelQwen3.8-27B-UD-Q4_K_XL.gguf · UD-Q4_K_XL · rev unsloth Dynamic V3.0 (file 17,559,178,144 bytes)
enginepwilkin/llama.cpp d3b5cc43d1fcfce891f2de94d5274ee40eceb21c · rocm · host aihydra (igpu)
flags-dev ROCm0 -ngl 999 -fa on -c 49152 -b 2048 -ub 2048 -ctk f16 -ctv f16 --parallel 1 --jinja --reasoning-effort low --spec-type draft-mtp --spec-draft-n-max 2 --metrics --port 5803 (isolation probe: runtime env toggled ENABLE_RETAINED_PM4=1 [run-0672] vs =0 [run-0673])
samplingtemperature 0 · max_tokens 256
templatenot recorded at test time
treefork — carrying con-0018, con-0019

Capability basis

unmeasured — performance numbers on this config stand on a guard alone, not an established capability

Runs on this config (2)

rundatekindsuitemetrics
run-06722026-09-06performancepm4-isolation@retained-ondecode_tps=20.17 · prefill_tps=262
run-06732026-09-06performancepm4-isolation@retained-offdecode_tps=19.45 · prefill_tps=264.6