Qwen3.5-122B-A10B(MTP)benchedguard 4/4
②Verdict
A 122-billion-parameter mixture-of-experts model activating 10B parameters per token, held at UD-Q4_K_M on a 128 GB unified-memory box. Flash attention is not optional on it at depth clm-0020, and quantised KV on the stock engine runs far below achievable speed at production context — one cherry-picked dequant commit recovers +70.3% clm-0022. On agentic work the model is correct more than it is fast; the 22-task score, turn medians and per-answer energy in the card above render straight from the record — five-task smokes could not resolve model differences clm-0036, which is why the protocol moved to n=22 clm-0037. Thinking mode is a net negative here: it deadlocks on hard tasks rather than degrading clm-0033. Serving it multi-slot destroys long-context retrieval while throughput stays healthy clm-0019 — run it thinking-off at --parallel 1. It also carries a reproducible rule-precedence bug: a permissive exception can short-circuit a hard precondition, 7 of 8 trials with an identical signature clm-0049 — agentic deployments want a hard-precondition guard in prompts or tooling.
③Best configuration
| model | Qwen3.5-122B-A10B-UD-Q4_K_M-00001-of-00003.gguf · UD-Q4_K_M |
| engine | ggml-org/llama.cpp 3653e6d · rocm · host aihydra (igpu) |
| flags | -ngl 999 -fa on -c 32768 --parallel 1 --load-mode none --jinja -rea off -ctk f16 -ctv f16 |
| template | not recorded at test time |
| tree | upstream — stock |
backfilled aged evidence — reconstructed from the archive · config record cfg-0025
carrying Production (aibeast) carries the slot-restore sidecar fork; the aihydra bench build above is stock. The KV-dequant patch is measured and adopt-worthy but not yet in the recommended flags — its rows are in the delta table below.
con-0001 clm-0022
every cell generated from the record at build time · throughput cells from cfg-0009 (same build 3653e6d, fa on, f16 KV) · provenance: measured-here throughout
③aDecode against context depth
③bEnergy per correct answer
④Other configurations tested — each as a delta against best
| variant | Δ decode | Δ turns (paired tasks) | Δ energy | note | records |
|---|---|---|---|---|---|
| draft-mtp speculation | +44% | — | — | Speculation reproduces the base model's distribution exactly (a neutral lever) — the ratio is measured against the same build's own no-speculation floor. | run-0007 clm-0003 |
| q8_0 KV · patched build | -0.5% @32k | +9.3% n=11 | -4.8% | The KV-dequant patch removes most of quantised KV's cost; a residual turn tax survives the patch on identical seeded tasks. | clm-0045 clm-0022 |
| q8_0 KV · stock build | -16% @32k | +39% n=9 | +63% | The mistuned path production actually ran — the patch exists because of this row. | clm-0022 clm-0038 clm-0041 |
| flash attention off | -10% @4k | — | — | The gap widens with depth, and quantised KV cannot create a context at all without flash attention. | clm-0020 clm-0028 |
| thinking ON hazard | — | deadlocks 2 of 5 tasks | Fails to terminate rather than degrading — the same tasks pass with thinking off. | clm-0033 | |
| --parallel 4 hazard | — | retrieval 0/8 against 8/8 at --parallel 1 | Throughput looks unchanged and healthy — no performance benchmark would see it. | clm-0019 | |
deltas computed at build time from the named runs, paired-task traces and metered windows — a hazard row shows words where a delta would mislead
⑤Open questions
- Wh per correct answer under the pinned independent simulator — the lab's headline unit. The current figure is a 5-task self-play smoke clm-0042 clm-0048.
- KL-divergence quality at depth — the instrument produced garbage on this build clm-0029; corpus pinned, overnight job queued.
- Cross-model τ² under the pinned simulator — current cross-model scores are provisional clm-0043 clm-0048.
- ROCmFP4 requant — a community build of this exact model reports +36% decode with MTP off clm-0009; a different runtime, so a new series, not a faster old one.
- Vulkan at depth — community reports agree Vulkan leads on decode at depth and disagree on magnitude clm-0044 clm-0031; the single-binary reproduction clm-0046 is not yet re-run under the τ² protocol.