Home › Evidence › Records › cfg-0132
⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.

cfg-0132

backfilled
citable URL: https://halobench.com/records/cfg-0132/ — this address never moves; the anchor /records/#cfg-0132 keeps resolving
modelQwen3.6-35B-A3B-MTP-UD-Q4_K_M.gguf · UD-Q4_K_M · rev 0b21525e972670ed59e1812e170b27c26355381f0656ecc4e25617ece7dac58b
engineggml-org/llama.cpp 7077abbe14c510cb829c93a1328c2815b5805ebd · ROCm0/gfx1151 · host aihydra (igpu)
flagsHO-009 plain/control fingerprint: /home/aihydra/src/llama.cpp-master-bailingmoe3, server build 50 commit 7077abb, ROCm0 Radeon 8060S, --load-mode none, -ngl 999, flash attention on, -b 2048, -ub 512, f16/f16 KV, --parallel 1, no proxy/llama-swap/ngram/external drafter/speculation.
templatenot recorded at test time
treeupstream — stock
aged evidence — reconstructed from the archive. Reviewer-admitted HO-009 plain baseline for the genuine Qwen3.6-35B-A3B-MTP UD-Q4_K_M artifact at /home/aihydra/bench-results/ho009-qwen36-35b-a3b-mtp-refinement-r2. This is an activation/safety/performance boundary only, not a capability or production claim.

Capability basis

measured on this config

Runs on this config (7)

rundatekindsuitemetrics
run-05242026-08-20performancestageA-c32768decode_tps=51.567783
run-05282026-08-20guardguard-c32768-depth8000tasks_passed=4 · tasks_total=4
run-05302026-08-20performancellama-bench@1decode_tps=52.429453 · prefill_tps=1061.377092
run-05312026-08-20performancellama-bench@1decode_tps=43.107742 · prefill_tps=600.282617
run-05322026-08-20performancellama-bench@1decode_tps=28.783536 · prefill_tps=251.893544
run-05332026-08-20performancellama-bench@1decode_tps=22.986826 · prefill_tps=175.137524
run-05342026-08-20performancellama-bench@1decode_tps=19.970905 · prefill_tps=143.241994