MERNIK — Measure-First Quantization Protocol

MERNIK ("the one who measures") is the evolution of the ASHQ1 battlefield zoo: fewer utility duels, more verdicts. The method (priority queue allocation) is built; MERNIK is how we prove and certify it.

Think of MERNIK as a finetune of ASHQ1: same base weights (queue, pins, tied groups), retrained objective (measure-first protocol, three-column verdicts), and expanded capabilities (slow capability ring, KLD-aware teachers, relief ceilings, the zoo bench, and norms shields).

1. Core Architecture & Evaluation Protocol

The Two Rings

MERNIK strictly separates allocation signals from capability certification:

Ring	Role	Trigger / Cadence
Fast / Allocator	Per-decision signal for the queue (imatrix, KLD-damage sweeps)	Every run
Slow / Capability	Scored finals, fixed seeds, exact capability metrics	Once per release

Rule: Allocator signals never certify. Capability scores never steer. Mixing them leads to the PPL disease.

Three-Column Verdicts

Every performance claim must present all three columns or stay unverified:

PPL (Canary): Cheap, demoted everywhere (often lies by sharpening).
KLD vs Ref (Rank Column): Used strictly for allocator decisions (Soulfate24 battery: fixed span, fixed chunks, Flash-Attention). Triage only — blind to long context, instruction drift, and tool formats.
Tasks (Verdict Column): The Slow Ring. The only column authorized to crown or kill a build.

Decision Rules & Signal Interpretation

Divergence: If metrics diverge, KLD wins allocator arguments; if they agree, trust the pair.
Silence is Signal: Empty completions scale with weakness across clean runs (Ox: 4–5 → Neo: 20–27 → Q2_K: 63 → S2: 115). Weak models go silent rather than babble. Decisiveness is governed by the finetune, not model size (e.g., on Qwen3.5 bones, Neo hesitates while OxCoder commits).
Saturation Rule: Base models at ~98% cannot resolve fine differences (97% vs 96%). Capability duels require bases at 60–85%, or they only answer the collapse threshold.

2. Quick Start & CLI Usage

Basic Execution

pip install -r requirements.txt

# 1. Preview tier allocation & dry-run (~1 sec)
python main.py --model M.gguf --imatrix M.imatrix.gguf --size 6800

# 2. Execute actual quantization
python main.py --model M.gguf --imatrix M.imatrix.gguf --size 6800 --run

Requires stock llama.cpp binaries (llama-quantize, llama-perplexity, llama-server). Custom paths via LLAMA_QUANTIZE, LLAMA_PPL, LLAMA_SERVER.

Full CLI Flags Reference

Flag	Description
`--model M.gguf`	Source weights (BF16/F16) — required
`--imatrix I.gguf`	Imatrix file; repeatable (`--imatrix A --imatrix B --imatrix-method max\|mean`)
`--size MIB`	Target file size in MiB (the primary budget knob)
`--output O.gguf`	Output path (default: `<model>-MERNIK.gguf`)
`--run`	Execute `llama-quantize` (without it: dry-run estimate only)
`--utility NAME`	Queue gain metric: `mse` (default) / `rmse` / `smape` / `logcosh` / `ssim` / `smape_ssim` / `smape_frag` / `mix` / `smse` (`balance` = deprecated alias) / `srmse`
`--mix-base NAME`	MIX utility: gain for non-king groups (`mse`/`rmse`/`smape`/`logcosh`)
`--mix-top-frac F`	MIX utility: fraction of top-importance king groups (default: 0.2)
`--top-down`	Reverse allocation: everything from F16, downgrade cheapest-loss-first
`--pin-norms`	Top-down native norms shield (norms stay F16, outside budget)
`--allow-q3-or-lower`	CAN_Q3 types (`ffn_gate/up/down`, `attn_output`, `ssm_out`) may drop to IQ2_XXS
`--relief labels.jsonl --relief-thr -0.5`	Groups below threshold get ceiling Q4; budget flows elsewhere
`--verify gpqa\|he\|all`	Slow-ring verify on fresh build or existing `--output` (busy GPU port aborts LOUDLY, never steals)
`--verify-tag TAG`	Result tag (default: from `--output` basename)
`--show-config` / `--show-floors`	Print tensor config / hard tier floors

Tier Auditing Tool

python scripts/audit_tiers.py --model M-6500.gguf
python scripts/audit_tiers.py --model M-6500.gguf --big 10
python scripts/audit_tiers.py --model M-6500.gguf --layer 31

3. Quantization Engine Mechanics

Queue & Allocation Principles

Every tensor starts at a floor tier based on its functional class. A global max-heap drains the target budget best-first by sum(importance) × Δ / MiB. Tied groups (identical imatrix energy) upgrade as a single unit with summed importance. Structural pins — MTP heads (→ Q8_0), output/token embeddings (→ Q5_K), MoE routers (→ F16) — sit outside the budget queue.

Utility Lenses: MSE vs SMAPE vs RMSE vs MIX vs SMSE vs SRMSE

MSE (default / spread): Relative-blind absolute gain. Tiers spread evenly (Q5/Q6-heavy middle). Best PPL/KLD on dense 9B.
SMAPE (barbell): Junk to the Q4 floor, kings to Q8 penthouses. Owns small budgets, loses big ones (budget law).
RMSE: Perfectionist rescale of MSE. Zoo: beats MSE on PPL twice. First verdict (RINIQ-M2-6500-RMSE-Q38): PPL 7.5411, GPQA 50.00%, HE 88.41% — takes recognition, loses verdict by 3pp to MSE. Budget law extended.
MIX (per-group): Kings by MSE, rest by --mix-base. Formulas live on different scales (mse Δ ~1e-3 vs smape ~1.2), so _scale() normalizes every formula to O(1) first — without it smape junk outbids mse kings 800:1 and the mix collapses. Dry-run geometry is new (wide Q5 spread, no Q8); verdict queued.
SMSE (blend 👑, ex-BALANCE): Geometric mean of the SMAPE-relative and MSE-absolute halves (both O(1)-normalized, MIX lesson). SMAPE's (ec+en) denominator silently cancels the sub-4 toxicity penalty; the blend carries real toxicity through its absolute half. First verdict (MiMo-4500, allow-q3): PPL 9.4092 vs SMAPE 10.5343 (−1.12), HE 54.88% vs 29.88% (+25pp) — rescue from the lava. Second verdict (MiMo-5100): PPL 8.7389 vs 8.7435 (tie), HE 71.34% vs 70.12% (+1.2pp), 3 empties — small-budget crown. Third verdict (MiMo-6500): PPL 8.7275 vs 8.7655 (tie), HE 72.56% vs 75.00% (−2.4pp) — LOSES big: at 6500 the middle decides (MSE Q6 67 vs SMSE 41), not penthouses (SMSE had more Q8, 125 vs 108). Verdict: SMSE owns small+medium budgets, MSE owns big. The bet behind it: any tier below Q4 hurts, and fewer is better.
SRMSE (SMAPE×RMSE, queued): Relative lens × COMPRESSED absolute (sqrt squeezes toxicity ×1.41 instead of ×2.0). Prediction: "SMAPE with manners" — PPL at/above SMAPE everywhere, HE between SMAPE and SMSE. Duel MiMo-5100-SRMSE running.

Distribution Comparison @ 6500 MiB (427 tensors, incl. 177 F16 norms/1D)

Tier      BU-MSE (spread)    BU-SMAPE (barbell)   TD-MSE       TD-SMAPE (F16 core)
F16       177 (2 MiB)        177 (2 MiB)          147 (2 MiB)  85 (949 MiB)
Q4_K      48 (972 MiB)       91 (1989 MiB)        71 (972)     236 (2368 MiB)
Q5_K      28 (1994 MiB)      3 (1345 MiB)         28 (1983)    7 (1411 MiB)
Q6_K      66 (2114 MiB)      7 (249 MiB)          71 (2127)    14 (486 MiB)
Q8_0      108 (1417 MiB)     149 (2913 MiB)       110 (1417)   85 (1287 MiB)

Takeaway: MSE fills the middle (Q5+Q6 = 94 tensors). SMAPE hollows it (10 tensors) to double the Q4 floor and gain 41 extra Q8 penthouses.

Norms Shield & Attention Core Insights

TD-MSE: downgrading ~30 small norm tensors to Q4 left PPL unaffected (7.7701 vs 7.7695) but dropped HumanEval by −3.7pp. Norms are free real estate for perplexity, but load-bearing walls for code. Norms Shield (TDN): native shielding recovers +1.2pp; post-hoc forcing disrupts greedy paths — shields must be native. Audit: RINIQ builds keep all 105 norm tensors intact F32.

4. Slow-Ring Protocol (HumanEval Execution)

Server (one model at a time, GPU is a strict queue):

llama-server -m MODEL.gguf --port 28082 -ngl 99 -c 8192 --jinja --log-disable

Sampling:

temperature 1.0, top_p 0.95, top_k 20, min_p 0.0,
presence_penalty 0.0, repetition_penalty 1.0, max_tokens 2048

presence_penalty 1.5 (vendor recipe) breaks thinking templates; 0.0 verified. Preflight /health before every battery; mid-run watchdog every 10 tasks (dead server aborts LOUDLY, never scores empties silently). Runner: scripts/run_humaneval.py (env HE_*, HUMANEVAL_OUT, HUMANEVAL_SERVER; HUMANEVAL_SERVE_MODEL = self-serve with auto-kill). Eval: human_eval.evaluation.evaluate_functional_correctness, pass@1, k=[1]. Results: eval_results/humaneval_<tag>.jsonl. Strictness upgrade: every battery is rescored with EvalPlus HumanEval+ (80× tests, CPU) — the HE+ column below.

5. Benchmark Results & Standings

The Budget Law

SMAPE owns small budgets; MSE owns big budgets.

Target Budget	MSE Utility	SMAPE Utility	Winning Lens
6500 MiB	PPL 7.7695 / HE 85.98%	PPL 7.8252 / HE 82.32%	MSE spread (+3.7pp)
5100 MiB	PPL 7.7962 / HE 79.88%	PPL 7.8012 / HE 82.93%	SMAPE barbell (+3.0pp)

The Sub-4 Law (confirmed 2026-09-23)

Any tier below Q4 hurts, and fewer is better. --allow-q3-or-lower at ~4 GB is lava in every family: MiMo-4500-SMAPE (PPL 10.53 / HE 29.88%) lands exactly where NeoHorse-3650-Q2 did (10.36 / 23.78%) — same physics, different bones. SMSE-4500 (9.41 / 54.88%, +25pp) proves limiting sub-4 exposure rescues the build. TOX (×2.0 effective-MSE penalty, still in constants.py) called it; SMSE proves it on the verdict column.

Reference Results I: NeoHorse-1-9B (official BF16 98.17% @ undisclosed ctx)

Build	File Size	PPL (ctx1024)	KLD vs Q8-proxy	HE pass@1	Key Finding
MERNIK-6500-MSE	6.83 GB	7.7695	0.0453	85.98% (141/164)	Outperforms stock Q6_K
MERNIK-6500-SMAPE	6.83 GB	7.8252	0.0476	82.32% (135/164)	Ties stock Q6_K
MERNIK-6500-TD-MSE	6.83 GB	7.7701	0.0452	82.32%	−3.7pp code loss vs MSE
MERNIK-6500-TDN-MSE	6.83 GB	7.7701	—	83.54%	Norms shield recovers +1.2pp
MERNIK-6500-TDN-SMAPE	6.83 GB	—	—	79.27%	Post-hoc shield disrupts greedy path
MERNIK-6500-TD-SMAPE	6.83 GB	7.8184	0.0505	82.32%	Holds Q6 level with 236 tensors at Q4
Q6_K (stock)	7.36 GB	7.9419	0.0118	82.32% (135/164)	Stock baseline
MERNIK-5100-SMAPE	5.36 GB	7.8012	0.0586	82.93% (136/164)	Q6-class code at −2.2 GB 👑
MERNIK-5100-MSE	5.36 GB	7.7962	0.0588	79.88% (131/164)	Budget law holds by 3pp
Q4_K_M (stock)	5.62 GB	7.7824	0.0875	76.83% (126/164)	Beaten by MERNIK-5100
Q2_K (stock)	3.83 GB	90.0284 💀	2.6354 💀	0.00%	Full collapse
MERNIK-3650-Q2	3.65 GB	10.3586	0.4706	23.78% (39/164)	Recovers sub-4-bit collapse

Reference Results II: OxCoder-9B (finetune duel, same Qwen3.5 bones)

Build	File Size	PPL (ctx1024)	HE pass@1	Note
Ox-MSE-6500	6.83 GB	7.5125	90.24% (148/164)	Distillate leads benchmark
Ox-SMAPE-5100	5.36 GB	7.5670	88.41% (145/164)	Compact build beats big NeoHorse
Ox-Q6_K (stock)	7.36 GB	7.6758	84.15% (138/164)	Allocation wins over stock flat

Full card: wepiqx/OxCoder-9B-MERNIK-GGUF.

Monster Standings: The RINIQ Series

Layer-fused models (OxCoder base + donor blocks; auto-maps via scripts/auto_fuse.py). Files: wepiqx/RINIQ-MERNIK-GGUF.

python scripts/fuse_layers.py --a OX --b ORN --c NEO --map "15:b,..."
python scripts/auto_fuse.py --a OX --b ORN --ia OX.imatrix --ib ORN.imatrix --rule rank-vote --out M7.gguf

Build	Size	PPL	GPQA-rec	HE pass@1	HE+ (EvalPlus 80×)	Empties
RINIQ-M2-MERNIK-5100-Q38 👑	5.0 GB	7.6130	47.47%	92.07% (151/164)	87.8 (−3.7)	8
RINIQ-M2-MERNIK-5100	5.0 GB	7.5819	50.00%	91.46% (150/164)	87.2 (−3.0)	6
RINIQ-M4a-MERNIK-5100	5.0 GB	7.5898	51.01%	91.46% (150/164)	87.2 (−4.3)	4
RINIQ-M1-MERNIK-5100	5.0 GB	7.6242	51.01%	90.85% (149/164)	85.4 (−4.8)	7
RINIQ-M4c-MERNIK-5100	5.0 GB	7.5863	47.98%	90.85% (149/164)	85.4 (−4.8)	6
RINIQ-M4b-MERNIK-5100	5.0 GB	7.5743	48.48%	89.63% (147/164)	84.1 (−4.9)	4
RINIQ-M3-MERNIK-5100	5.0 GB	7.7111	50.00%	86.59% (142/164)	83.5 (−3.1)	13
RINIQ-M2-SMAPE-6500	6.8 GB	7.5730	48.48%	89.63% (147/164)	83.5 (−5.5)	5
RINIQ-M2-MSE-6500	6.8 GB	7.5263	48.48%	91.46% (150/164)	85.4 (−4.8)	8
RINIQ-M5-MERNIK-5100	5.0 GB	7.6587	52.02%	91.46% (150/164)	84.8 (−6.1)	7
RINIQ-M6-MERNIK-5100	5.6 GB	8.5137	46.97%	83.54% (137/164)	79.9 (−3.6)	23
RINIQ-M7-MERNIK-5100	5.0 GB	7.6033	49.49%	88.41% (145/164)	86.0 (−2.4)	11 — first auto-fused: mid-pack, method validated
RINIQ-M2-MERNIK-5100-MIX	5.0 GB	7.6001	48.99%	89.63% (147/164)	85.4 (−3.6)	6 — novel geometry, mid-pack
RINIQ-M2-RMSE-Q38-6500	6.8 GB	7.5411	50.00%	88.41% (145/164)	82.9 (−4.9)	6 — RMSE's first verdict: takes GPQA, loses HE by 3pp

Sub-block donor mapping: M1 = Ox + Orn 15,19,23,27 + Neo 31. M2 = Ox + Orn 24,25,26 + Neo 31. M3 = M1 + blks 0–8 soup. M4a = M2 + 16←Orn. M4b = M2 with 25←Ox. M4c = M2 + 30←Orn. M5 = M1∪M2 (Orn 15,19,23,24,25,26,27). M6 = Orn base + Ox 16,24,25,26,31 (mirror, BASE RULES). M7 = auto rank-vote Orn{10,14,15,16,18}.

External Reference: JackOD-9B-Coder (different harness!)

Corporate 4-way omnimerge_v2 over the same Qwen3.5-9B ancestor. Quoted from their README — greedy temp 0.0, mixed banks (their own † admits it). Same-harness duel (LiveCodeBench v6 55 hard) queued.

Build	HE	HE+	LCB v6 55 hard
RINIQ-M2-MERNIK-5100-Q38 (ours)	92.07	87.8	⏳
JackOD-9B-Coder (Q6_K, theirs)	88.41	82.32	78.18

Fast-Ring Diagnostic: GPQA-Recognition

Recognition decisiveness (first-token P(letter) over top-30, no CoT). Via scripts/gpqa_duel.py.

Build	GPQA-rec	HE pass@1	Note
Neo-MSE-6500	51.5%	85.98%	Thinker spreads first-token mass
Neo-SMAPE-5100	48.0%	82.93%	Compact barbell
Ox-MSE-6500	49.5%	90.24%	Decisive stabs the letter
Ox-SMAPE-5100	49.0%	88.41%	High task execution

6. Advanced Experiments

MTP Head Grafting

Transplant of a 15-tensor (464 MB) MTP donor head onto OxCoder-9B via scripts/graft_mtp3.py. Draft engages (5–23% acceptance); economics negative (31→20 t/s on mismatched head). Mechanics proven; full saga in SAGA.md.

Teacher & Gini (v3 DONE 2026-09-19)

Damage-sweep labels (scripts/group_damage_sweep.py) train the Gini allocator. v3 complete: 168/168 labels (Q5_K→Q2_K drops, MiniCPM5-2B, forge): damage −0.046…+0.515, mean 0.070, Gini 0.364, only 3 reliefs. Late units carry the signal (ffn_gate@41 +0.51, ffn_down@41 +0.33, ffn@39/40) + early attn@2. vs older teacher (Gini 0.498, max 0.173): bigger drops hit harder (max 3×) but spread wider (Gini lower) — signal with teeth, less concentrated. MLP un-paused: retrain Gnom on v3 → netdmg duel. Gini answers to Soulfate24.

KLD Battery & Relief

KLD-vs-ref is the rank column (Soulfate24 battery). Measured relief ceilings (--relief) pin sweet-spot groups at Q4.

7. Supported Architectures & Multi-Scale Scaling

Arch	Detection	Key Features
`qwen35`	SSM + QKV	Hybrid attention, SSM layers, GQA, MTP head pinning
`mellum2`	MoE (`exps` tensors)	Mixture of Experts, GQA, routers pinned F16
`bailingmoe3`	KDA+MLA + MoE	Ling-3.0 family: routed + shared experts
`granite`	`general.architecture`	Dense GQA, 40 layers, split Q/K/V
`spark2_5`	`general.architecture`	Dense + hybrid sliding-window attention (1 full + 3 SWA)
`gemma4`	layer-scale norms	QAT support, Q4_K attention floor
llama (generic)	tensor names	Dense GQA (MiniCPM5, NeoHorse, etc.)

Scaling: Spark-1.7B @1000 (zoo stand) → Spark-4B @4000 (PPL 31.25 vs Q8 30.86) → Ornith-9B-MTP @6500 (PPL 8.6541) → NeoHorse-9B @6500 (HE 85.98% > Q6_K 82.32%).

8. Ecosystem & Repositories

Lineage: Apache-2.0. Built with Soulfate24, whose KLD battery ended our PPL theater. Field journals with every scar: FUSION.md (monsters), README-ZOO.md (utility duels), SAGA.md (MTP graft).

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including wepiqx/MERNIK

MERNIK

Collection

Measure-first quantization protocol • 1 item • Updated 12 days ago • 1