← back to catalog · registered 2026-11-12 14:03

kabcnd/Kenzan-Flint-40B-Abliterated-Hepta48Up-GGUF

kabcnd 40B GGUF MoE second-order
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/kabcnd%2FKenzan-Flint-40B-Abliterated-Hepta48Up-GGUF"
Response includes
  • classification m8
  • files 3
  • author_summary 3 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M8
Primary method

Repackaging (quantization)

Applied on top of direct removal inherited from the base model.
Confidence
MEDIUM
Inherited from base model
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=1
  • assume M1 (base ablation) + M8 (GGUF quant) - default when producer unknown
Refusal direction extraction

No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.

What is a refusal direction? →
Downloads · 30-day
60
Likes
0
Model age
1w ago
created 2026-11-12
Downloads over time
Now 60 → from 0 ↑ 0%
0224466 0 on Nov 12 60 on Nov 22 Nov
Nov 12 → Nov 22 · 11 snapshots · spans 10 days

Genealogy 0 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

Metadata

License
apache-2.0
Languages
en ru zh
Tags
gguf kenzan ternary moe 40b abliterated uncensored apple-silicon metal hadamard ptq1_0 q4_0

Related

Total size
9.77 GB
Files
3
Quantizations
1
Registered
2026-11-12 14:03
Last updated on HF
2026-11-18 06:18

Files by quantization

Auxiliary files 3 files 9.77 GB
Kenzan-Flint-40B-Abliterated-Hepta48Up-Q4_0-M6-16G.gguf 9.77 GB ******** download
README.md 9.38 KB ae9f5f62 download
.gitattributes 1.57 KB ed4189b4 download

README current version from Hugging Face


license: apache-2.0
base_model: orcarouter/Flint-abliterated-GGUF
tags:

  • kenzan
  • gguf
  • ternary
  • moe
  • 40b
  • abliterated
  • uncensored
  • apple-silicon
  • metal
  • hadamard
  • ptq1_0
  • q4_0
  • speculative-decoding
  • kenzan-router
    pipeline_tag: text-generation
    language:
  • en
  • ru
  • zh
    inference: false
    extra_gated_heading: "Mandatory Liability Waiver & Uncensored Model Agreement"
    extra_gated_prompt: >
    WARNING: This repository hosts an ABLITERATED (uncensored) 40B MoE language model with safety refusals removed.
    By requesting access to, downloading, or using this model (or any other models published by kabcnd / the Kenzan Project),
    you explicitly acknowledge, sign, and agree to the following legally binding terms:
    (1) SOLE RESPONSIBILITY: You assume 100% personal and legal responsibility for all prompts, outputs, deployments, and consequences arising from the use of this model or any other models provided by kabcnd / the Kenzan Project.
    (2) FULL RELEASE OF LIABILITY: The author(s), maintainer(s), and contributors (kabcnd / Kenzan Project) bear ZERO liability and are hereby completely released and held harmless from any claims, damages, misuse, illegal acts, or direct/indirect harm caused by this model or any other Kenzan / Huihui / Flint models.
    (3) LAWFUL USE ONLY: You will use this model strictly for lawful research, red-teaming, security evaluation, or authorized local testing in compliance with all applicable laws in your jurisdiction.
    extra_gated_fields:
    Full Name or Signature (required to sign waiver): text
    Organization / Affiliation (or Independent Researcher): text
    I confirm that I am at least 18 years old: checkbox
    I accept 100% personal responsibility and release kabcnd / Kenzan Project from all liability for any harm caused by this or other models: checkbox
    extra_gated_button_content: "Sign Liability Waiver & Unlock Download"

☕ Support Local AI Development: Help me acquire a Mac mini 24GB/32GB to optimize and test local MoE models up to 80B parameters. Donate via Ko-fi →

[!CAUTION]
Kenzan-Flint-40B-Abliterated-Hepta48Up — "40B Uncensored MoE, 9.77 GiB, 79 t/s, Zero Guardrails."
We took an 11.17 GiB abliterated 40B MoE, rotated 141 expert tensors through FWHT_512 Hadamard surgery, pinned the Hepta48Up tail grid in Q4_0, fixed the hidden expert_weights_scale trap, and squeezed the whole thing into 9.77 GiB so a base 16 GB Mac mini can sprint at nearly 80 tokens/sec without a single refusal.


Kenzan-Flint-40B-Abliterated-Hepta48Up (9.77 GiB GGUF)

Kenzan-Flint-40B-Abliterated-Hepta48Up (Kenzan-Flint-40B-Abliterated-Hepta48Up-Q4_0-M6-16G.gguf) is a hybrid ternary-4bit Mixture-of-Experts (MoE) uncensored language model (40B total parameters, 256 routed experts + 2 shared experts, 49 layers) distilled and compressed from orcarouter/Flint-abliterated-GGUF (Flint-abliterated-Mix-12G.gguf, 11.17 GiB) specifically for Apple Silicon 16 GB Unified Memory (Mac mini M6).

Short Summary: Kenzan-Flint-40B-Abliterated-Hepta48Up (9.77 GiB GGUF) — Hybrid PTQ1_0 (1.75 bpw) + Q4_0+FWHT_512 40B uncensored MoE tuned for Apple Silicon 16 GB Unified Memory, achieving 29.38 WikiText-2 holdout PPL, 69.7–79.4 t/s generation, and 262k context at 11.82 GiB RSS.


Architecture & Pin-Grid Surgery

  • Ternary Crown (PTQ1_0, 1.75 bpw): All 133 non-anchor routed expert tensors (blk.2..blk.48) are dequantized from Flint-abliterated-Mix-12G.gguf, rotated via block-512 Fast Walsh-Hadamard Transform (FWHT_512), quantized with flint_imatrix.dat force-density (0.691 active trits), and stabilized with permanent Salient Clipping (κ = 0.15, η = -0.025).
  • Pin Grid (Hepta48Up): down_exps anchored at layers {8, 12, 20, 40, 44, 46, 48} + output-adjacent blk.48.ffn_up_exps in Q4_0 + FWHT_512.
  • Scale Alignment (flint.expert_weights_scale = 1.0): Aligned GGUF expert scaling to match the abliterated donor (1.0 instead of 1.24).
  • Always-On Backbone: High-precision imatrix-calibrated Q8_0 token_embd, Q6_K output.weight, Q5_K/IQ4_XS/IQ3_XXS attention and shared experts.

Verified Benchmarks (Mac mini M6, 16 GB Unified Memory)

Measured directly on Apple Silicon Metal GPU (-ngl 99 --no-host -fa on -c 262144 --cache-type-k q4_0 --cache-type-v q4_0) with ngram-map-k (3x4) speculative decoding:

Metric Flint-abliterated-Mix-12G (Donor) Kenzan-Flint-40B-Abliterated-Hepta48Up
GGUF File Size 11.168 GiB 9.768 GiB (-1.40 GiB)
Process RSS (-c 262144) N/A 11.82 GiB (12,392,128 KB)
WikiText-2 Holdout (c512) PPL 28.1418 ± 0.9639 29.3767 ± 0.8194
Generation Throughput (tg) — 69.7 – 79.4 t/s (live router 11434)
Prompt Ingestion (pp) — 190.1 – 275.2 t/s

Quick Start & Recommended Deployment (Apple Silicon 16 GB)

Important Runtime Requirement: Stock upstream llama.cpp cannot load Kenzan GGUF files because they use packed ternary PTQ1_0 weights and prism.hadamard.* phase-rotation metadata. You must use the patched PrismML-Eng/llama.cpp runtime (mac-flint v0.7.0-dev build 12128 or newer).

1. Required Binaries & Patched Metal Libraries

Download the model (after signing the gated waiver above), the precompiled macOS ARM64 runtime bundle, and tools/kenzan_router.py:

hf download kabcnd/Kenzan-Flint-40B-Abliterated-Hepta48Up-GGUF \
  Kenzan-Flint-40B-Abliterated-Hepta48Up-Q4_0-M6-16G.gguf \
  bin/llama-flint-metal-macos-arm64.tar.gz \
  tools/kenzan_router.py --local-dir .

tar -zxvf bin/llama-flint-metal-macos-arm64.tar.gz
export DYLD_LIBRARY_PATH="$(pwd)/llama-flint:$DYLD_LIBRARY_PATH"

2. Mandatory Metal Environment Variables

export GGML_METAL_PTQ1_MULTICOL=1
export GGML_METAL_FUSE_GATHER_MV=1
export GGML_METAL_TENSOR_ENABLE=1
export GGML_METAL_SHARED_BUFFERS_ENABLE=1
export GGML_METAL_Q1_0_POPCNT=1

3. Production llama-server Launch Command (16 GB Safe, 262k Context)

./llama-flint/llama-server \
  -m Kenzan-Flint-40B-Abliterated-Hepta48Up-Q4_0-M6-16G.gguf \
  --host 127.0.0.1 --port 11435 \
  -ngl 99 --no-host -fa on \
  -c 262144 --cache-type-k q4_0 --cache-type-v q4_0 -np 1 -kvu \
  -t 4 --threads-draft 2 -b 512 -ub 512 --ctx-checkpoints 8 --cache-ram 2048 \
  --jinja --metrics \
  --spec-type ngram-map-k --spec-ngram-map-k-size-n 3 --spec-ngram-map-k-size-m 4 \
  --spec-draft-n-max 3 --spec-draft-n-min 1 \
  --samplers "top_k;dry;penalties;top_p;min_p;temperature" \
  --temp 0.20 --top-p 0.95 --top-k 40 --min-p 0.08 \
  --repeat-penalty 1.05 --repeat-last-n 256 \
  --dry-multiplier 0.8 --dry-base 1.75 --dry-allowed-length 3 --dry-penalty-last-n 512 --dry-sequence-breaker $'\f' \
  --reasoning on --reasoning-format deepseek --reasoning-budget -1 --reasoning-preserve \
  --alias "kenzan-flint:40b-abliterated"

You can also run the unified multi-model proxy tools/kenzan_router.py on port 11434 (kenzan-flint:40b-abliterated, kenzan-flint:40b, kenzan-qwen3.8:27b, bonsai-2:latest).


Liability Waiver & Responsible Use Disclaimer

This model is an abliterated (uncensored) research artifact with safety refusal directions removed. By downloading, loading, or executing this model or any other model published by kabcnd / Kenzan Project, you agree that:

  1. You assume full and exclusive responsibility for all outputs, deployments, and consequences.
  2. The authors and maintainers (kabcnd / Kenzan Project) are completely released from any and all liability for misuse, damages, or harm arising from this or any other model in the Kenzan / Huihui / Flint series.
  3. You will use this model solely for lawful, ethical, and authorized research purposes.

Acknowledgements & Inspiration

  • Prism ML (https://huggingface.co/prism-ml) — Inspired by and built upon the Bonsai-2 architecture, PTQ1_0 ternary quantization, FWHT_512 Hadamard transforms, and Metal kernel foundations.
  • OrcaRouter (https://huggingface.co/orcarouter/Flint-abliterated-GGUF) — For the upstream Flint-abliterated-Mix-12G donor model.
  • Misha — Special thanks to Misha for ongoing support, encouragement, and financial assistance that helped keep this local AI hardware research moving forward.

Support the Project & Funding Goals

I am an independent researcher optimizing frontier 27B–80B models to run locally on affordable consumer hardware without expensive NVIDIA server clusters.

Current Hardware Fundraising Goal: Mac mini 24 GB / 32 GB Unified Memory

  • Why it matters: Testing and calibrating 40B–80B MoE models on a base 16 GB Mac mini hits a hard physical ceiling (~12.5 GB wired Metal limit). A Mac mini 24 GB / 32 GB will unlock higher-precision anchor grids, larger KV caches, and full 70B–80B ternary/4-bit hybrid models for the open-source community.
  • What your support funds: Direct hardware upgrades, open-source GGUF releases, Metal kernel benchmarks, and reproducible quantization tooling.

☕ Support Local AI Research on Ko-fi →

Even a small contribution or sharing this repository helps enormously. Thank you! ❤️

Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration