Get the app
← back to catalog · registered 2026-08-22 13:56

SC117/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-GGUF

SC117 Qwen 35B GGUF MoE multimodal 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/SC117%2FQwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-GGUF"
Response includes
  • classification m3
  • files 8
  • hub_downloads_all_time 301,366
  • author_summary 18 models
  • readme_text full
Try next
GET/models?author=SC117
All 18 models by SC117
 GET/models?family=qwen&method=m3
Sibling qwen models · same method
 GET/methods/m3
What "m3" means · taxonomy
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M3
Primary method

Layer-wise ablation

Applied on top of direct removal inherited from the base model.
Confidence
HIGH
Inherited from base model
M1 Direct removal →
Why this label 2 signals
Producer identity confirmed by naming conventions, tags or the model card. This label is very unlikely to change.
  • 'heretic' in model name (Heretic-produced)
  • Heretic uses layer-wise optimization (M3) with underlying direction removal (M1)
How we label → About M3 →
Refusal direction extracted via
Extraction technique

Difference-of-means

FromArditi et al. (2024) ↗
Confidence
HIGH
Why we say so
name contains 'heretic'; Heretic default extraction is difference-of-means (Arditi 2024)
What is a refusal direction? → Read the paper ↗
Downloads · lifetime
301K
71K last 30d - stable
HF trending score
7
mild recent interest
Likes
144
Model age
4mo ago
created 2026-05-26
Downloads over time
Now321.6K→from0↑0%
0117.9K235.8K353.7K0 on May 26321.6K on Oct 12MayJunJulAugSepOct
May 26 → Oct 12 · 61 snapshots · spans 139 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.

llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved↓ 749 this lineage
SC117/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-GGUF↓ 70,532

Metadata

License
apache-2.0
Tags
transformers gguf qwen3_5_moe qwen3_5 heretic uncensored decensored abliterated mpoa mtp apex quantization

Related

Other by SC117
Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF 907K Swift-1.5-Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF 106K Spark-X2.5-4B-abliterated-FIT-GGUF 24K occamy-1.0-abliterated-FIT-GGUF 18K Qwen3.8-27B-Uncensored-FIT-GGUF 15K
Same base model
llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-GGUF 31K llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-GPTQ-Int4 19K mradermacher/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-i1-GGUF 4K mradermacher/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-GGUF 2K llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-NVFP4-Experts-Only-GGUF 1K
Total size
75.2 GB
Files
8
Quantizations
2
Registered
2026-08-22 13:56
Last updated on HF
2026-10-03 10:22
Source huggingface.co/SC117/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-GGUF ↗ Base model llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved →

Files by quantization

F16 1 file 858 MB
mmproj-Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-F16.gguf 858 MB c8e70234 download
Auxiliary files 7 files 75.2 GB
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-I-Balanced.gguf 24.2 GB 782636c4 download
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-I-Quality.gguf 21.9 GB 31a6a415 download
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-I-Compact.gguf 15.8 GB ff82d199 download
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-I-MINI.gguf 13.3 GB e4eae0e2 download
README_zh.md 21.4 KB a18f3d42 download
README.md 21.3 KB 5da6e4b4 download
.gitattributes 2.03 KB a296e3d6 download

README current version from Hugging Face

library_name: transformers
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen3.6-35B-A3B/blob/main/LICENSE
pipeline_tag: image-text-to-text
tags:

  • qwen3_5_moe
  • qwen3_5
  • heretic
  • uncensored
  • decensored
  • abliterated
  • mpoa
  • mtp
  • apex
  • quantization
    base_model:
  • llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved
APEX MTP Uncensored

Qwen3.6-35B-A3B

English | 📖 中文文档

🔄 Update (2026-06-26)

Optimized quantization strategy with improved imatrix calibration. File sizes reduced while maintaining quality:

  • I-Compact: 16.61 GB → 15.85 GB (-0.76 GB)
  • I-Balanced: 24.24 GB → 24.18 GB (-0.06 GB)
  • I-Quality: 21.90 GB → 21.90 GB (unchanged)
💡 What is APEX?

These GGUF files are quantized using APEX, a novel MoE-aware mixed-precision quantization technique that outperforms standard quantization methods while being significantly smaller.

APEX beats Q8_0 perplexity at half the size — and even beats F16.

APEX classifies every tensor by its role — routed expert, shared expert, or attention — and applies a layer-wise precision gradient, giving the most sensitive edge layers higher precision and compressing the redundant middle layers more aggressively.

📦 APEX Quantization Tiers
FileSizeProfileBest For
*-APEX-I-Quality.gguf21.90 GBI-QualityHighest quality, best accuracy
*-APEX-I-Balanced.gguf24.18 GBI-BalancedBest all-rounder, recommended
*-APEX-I-Compact.gguf15.85 GBI-CompactBest quality/size ratio
📊 Why APEX?
MethodSizePerplexityHellaSwagSpeed
F1664.6 GB6.53782.5%30.4 t/s
Q8_034.4 GB6.53383.0%52.5 t/s
APEX I-Quality21.3 GB6.55283.5%63.1 t/s
APEX I-Balanced23.6 GB6.54883.0%61.4 t/s
APEX I-Compact16.1 GB6.66981.8%69.8 t/s
APEX Mini12.2 GB7.08881.0%74.4 t/s

Benchmarks on Qwen3.5-35B-A3B, NVIDIA DGX Spark (GB10, 122 GB VRAM).

📐 I-Variant: Diverse Imatrix Calibration

I-variants use a diverse calibration dataset spanning chat, code, reasoning, and tool-calling — no Wikipedia. This produces higher accuracy on real-world benchmarks, lower KL divergence, and tiny perplexity increase on wikitext.

🔓 Original Model

Base: llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved

Decensored version of Qwen/Qwen3.6-35B-A3B, made using Heretic v1.3.0 + MPOA.

88% fewer refusals (10/100 Uncensored vs 83/100 Original) while preserving model quality (0.0015 KL divergence).

⚡ MTP Speculative Decoding

All 20 MTP layers preserved. Use --spec-type draft-mtp --spec-draft-n-max 2 for 20-50% speed boost.

🧠 Model Details
ArchitectureQwen3.5 MoE (Mixture of Experts)
Parameters35B total, 3B active per token
Experts256 routed experts, 8 active per token
Layers40 transformer layers + 1 MTP layer
Context262,144 tokens
DecensoringHeretic v1.3.0 + MPOA (88% fewer refusals, 0.0015 KL)
🚀 Usage

llama.cpp

hf download SC117/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-GGUF --include "*.gguf" --local-dir ./models ./llama-server -m ./models/...APEX-I-Compact.gguf -ngl 99 -ncmoe 16 -c 131072 --spec-type draft-mtp --spec-draft-n-max 2

Ollama

echo 'FROM ./...APEX-I-Compact.gguf' > Modelfile ollama create my-model -f Modelfile && ollama run my-model

🎛️ Recommended Settings
ModeParameters
Thinking (general)temp=1.0, top_p=0.95, top_k=20, pres_penalty=1.5
Thinking (coding)temp=0.6, top_p=0.95, top_k=20
Instruct (general)temp=0.7, top_p=0.8, top_k=20, pres_penalty=1.5
Instruct (reasoning)temp=1.0, top_p=1.0, top_k=40, pres_penalty=2.0

Links

Citation

@misc{qwen36_35b_a3b,
title = {{Qwen3.6-35B-A3B}: Agentic Coding Power, Now Open to All},
url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b},
author = {{Qwen Team}},
month = {April},
year = {2026}
}

README history 20 versions

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-10-03Add optional Ko-fi support banner313e55f21.5 KB
    Loading...
  2. 2026-06-26Update documentation495ab5b21.1 KB
    Loading...
  3. 2026-05-31fix: use full URL for language switchera1922f420.3 KB
    Loading...
  4. 2026-05-31docs: add Chinese README70d347c20.2 KB
    Loading...
  5. 2026-05-30Upload README.md with huggingface_hub622508020.1 KB
    Loading...
  6. 2026-05-30Upload README.md with huggingface_hub17de15619.2 KB
    Loading...
  7. 2026-05-30Upload README.md with huggingface_hubb3a78b919.9 KB
    Loading...
  8. 2026-05-30Upload README.md with huggingface_hubeb48cdd21.4 KB
    Loading...
  9. 2026-05-30Upload README.md with huggingface_hub596d18a20.6 KB
    Loading...
  10. 2026-05-30Upload README.md with huggingface_hub4299b5520.5 KB
    Loading...
  11. 2026-05-30Upload README.md with huggingface_hub41fd07019.5 KB
    Loading...
  12. 2026-05-30Upload README.md with huggingface_hub00e770c20 KB
    Loading...
  13. 2026-05-30Upload README.md with huggingface_hub361cc8819 KB
    Loading...
  14. 2026-05-30Upload README.md with huggingface_hub6bb2b4318.9 KB
    Loading...
  15. 2026-05-30Upload README.md with huggingface_hub9b0b1838.4 KB
    Loading...
  16. 2026-05-30Upload README.md with huggingface_hubf3855658.4 KB
    Loading...
  17. 2026-05-30Upload README.md with huggingface_hub95d550c8.2 KB
    Loading...
  18. 2026-05-30Upload README.md with huggingface_hubb59d8266.3 KB
    Loading...
  19. 2026-05-30Upload README.md with huggingface_hub4353cf59.1 KB
    Loading...
  20. 2026-05-30Upload README.md with huggingface_hub75f344710.6 KB
    Loading...

Discussions 5 threads

  1. 2026-08-17Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved-APEX-I-Balanced.gguf错误open4 💬#5
    Loading...
  2. 2026-08-02This quant is magical!open6 💬#4
    Loading...
  3. 2026-07-28Mini GGUFopen3 💬#3
    Loading...
  4. 2026-07-19Excellent performance!open1 💬#2
    Loading...
  5. 2026-07-09APEX Hereticopen2 💬#1
    Loading...

Continue on Hugging Face →

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.

Get the app Browse the catalog
Open in Abliteration