Nex-N2.5-mini — Sangreal ICE

Four ICE 1.6 tiers of nex-agi's Nex-N2.5-mini (35B total, 3B active, Qwen3.5-MoE architecture), quantized against Sangreal — a purpose-built 77-bucket calibration corpus — with an importance matrix measured on Nex itself, and shipping Qwen-Sharp v22.5.0-arrow4.0 — our agent-tuned chat template, a heavily modified derivative of peculiar-ragdoll's Qwen-Sharp (spec below). Vision works through Nex's own projector, mmproj-Nex-N2.5-mini-Q8_0.gguf.

At a glance — against bartowski at the same size

Mean KLD to BF16, one session. Size-matched = bartowski's imatrix ladder for the same model, interpolated (log-linear) at each ICE file's exact byte size; negative = ICE closer to BF16. Paired = the same 64 chunks scored for both files against the same reference.

Code at 16K context (code.test.raw, n_ctx 16384 × 8), mean KLD to BF16 vs file size; lower-left is better. Raw numbers in Table 2 below and in measurements/2026-10-03/.

Demo — Coral Dragon Reef Rescue

A complete browser game — one self-contained index.html, no assets, no network calls — generated in a single shot by the sibling OrcaRouter build of this quant, Nex-N2.5-mini-OrcaRouter-Sangreal-23G-ICE.gguf, in about 45 minutes, running locally in the Nous Hermes desktop harness on Windows 11 with an RTX 3070 (8 GB VRAM), 32 GB RAM and no MTP.

Play: Hugging Face Space · GitHub Pages · Source

Measurements

One binary (llama.cpp 3653e6d), one reference (Nex-N2.5-mini BF16), one session (2026-10-03, RTX PRO 6000 96 GB). 64 chunks at n_ctx 2048; the long-context table uses 8 chunks at n_ctx 16384. Every file below was measured in this session. Raw logs: measurements/2026-10-03/.

Table 1 — Code (code.test.raw)

PPL(base) 2.322497 · BF16 = 100

Table 2 — Code at long context (code.test.raw, n_ctx 16384 × 8)

Compare rows within this table only.

PPL(base) 2.385381 · BF16 = 100

Table 3 — English text (WikiText-2)

PPL(base) 6.509631 · BF16 = 100

overall = 0.70/(1 + meanKLD) + 0.30 * sameTop1, ×100. BF16 = 100. The same composite as the Ornith-1.5 ICE card: 70% on how close the whole output distribution stays, 30% on agreement about the argmax.

Read the tail columns as shape, not order. 99.9% KLD is roughly the 33rd-worst token of 32,768 — an extreme order statistic with large sampling variance, so it inverts between adjacent files without meaning anything. Rank on mean KLD.

Two bpw columns. Only 8 of 256 experts fire per token, so a bit in ffn_*_exps is worth ~3% of a bit in attention, the shared expert or the output head. active bpw weights by that (token embedding excluded: it is a lookup, not a matmul); file bpw is just size ÷ parameters.

Don't compare the tables to each other. Code is more predictable text, so every file scores about half the divergence on it. Compare rows within a table.

Every row is measured against Nex-N2.5-mini's own BF16 — bartowski's unmodified BF16 conversion of nex-agi's weights (its tensor set, types and metadata match our own conversion plan tensor for tensor). Quantizations of other models are deliberately absent: scoring them here would measure the distance between models and call it quantization damage.

What Sangreal is

The calibration corpus these tiers were quantized against. Built from primary sources, not assembled from eaddario's set, bartowski's calibration_datav5, or any other ready-made calibration file. 77 buckets, 9,533 documents, 47,443,549 tokens, sha256 85a6b823a6762bf6….

The full render is 92,663 chunks. These tiers use a 10,000-chunk pass over it, measured on Nex-N2.5-mini itself - 17x the 573-chunk imatrix behind the stock quants (bartowski's calibration-v6).

Split: technical 70.2 · science+domain 20.1 · language 9.7.

Synthetic is 21.7% by bytes and sits in exactly three buckets — teacher:k2-horizon, teacher:deepseek-v4-pro, teacher:hermes. Everything outside reasoning and agentic is real text.

Language — 11 languages, 5 non-Latin scripts

The spread across the whole block is 0.06 points, and that is deliberate: language's job here is activation, not ranking. Each bucket sits just above the saturation floor at 44.9% consumption, so these buckets select for the first time — the legacy v1 spec had consumed 96% of turkish and 92% of english.

Mathematics — absent from the pool until this cycle

Sourced from the LaTeX algebraic geometry, mathlib4, UniMath, math-comp, open-web-math, AutoMathText and proof-pile.

The imatrix and the recipes are not published with this release.

What's inside

ICE = Isolation of Compounding Error: allocate bits by how far a quantization error travels, not by how large the activations are. Routers stay F32, SSM decay gates stay F32, the KV-cached projections stay F16, the always-on dense path stays Q8_0, and the budget goes to the routed expert bank — most of the parameters, but only 8-of-256 active per token. ICE 1.6 keeps token_embd at Q8_0 (its error propagates with context) and moves the output head to Q6_K (its error stays flat from 2K to 16K). Method and the cases where it does not win: gbuzhf/ICE-quantization.

No MTP head. nex-agi's config declares one MTP layer, but the checkpoint carries no MTP weights. These tiers are the 40-block trunk; there is no built-in draft model.

Chat template: Qwen-Sharp v22.5.0-arrow4.0 (ours; spec below). Nex-N2.5-mini uses Qwen3.6's tokenizer unchanged (vocabulary, merges and special tokens; seven extra audio/TTS tokens), so the template's delimiters are native tokens — checked on every tier at signing.

Vision. Nex retrained its vision tower: every vision tensor we sampled differs from Qwen3.6's, projector included. This repo therefore ships Nex's own projector, converted from nex-agi's weights to Q8_0 (334 tensors). Load it with --mmproj mmproj-Nex-N2.5-mini-Q8_0.gguf.

Chat template — Qwen-Sharp v22.5.0-arrow4.0

The GGUFs still embed arrow.3.1; pass the file shipped here to use arrow4.0:

llama-server -m <model>.gguf --jinja --chat-template-file chat_template_sharp-v22.5.0-arrow4.0-nex.jinja

arrow4.0 is our agent layer on Qwen-Sharp v22.5.0. Plain chat stays byte-identical to Sharp; with tools it adds short agent rules, one note per tool result, a STATUS line, thinking only at decision points, .arrow project memory and a hand-back report when work stalls. The -nex file opens replies the Nex way, so the harness's reasoning level is respected (default medium = adaptive).

Full description and switches: gbuzhf/Qwen-Sharp-Arrow-Chat-Templates.

Files

• measurements/2026-10-03/ — raw llama-perplexity output for every row of every table (4 ICE tiers + 7 bartowski files × 3 columns), the three BF16 base runs, and the session summary.

Credits

KLD measures fidelity to this repo's reference and nothing else — not reasoning, tool use or speed. It is also within-model: every file here is measured against Nex-N2.5-mini's own BF16, so these values are not comparable to another repo's. Compare across models with PPL.