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

catplusplus/Qwen3.6-27B-uncensored-heretic-v2-NVFP4-MTP

catplusplus Qwen 9.4B
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/catplusplus%2FQwen3.6-27B-uncensored-heretic-v2-NVFP4-MTP"
Response includes
  • classification m3
  • files 11
  • benchmarks 11 entries
  • hub_downloads_all_time 1,276
  • author_summary 3 models
  • readme_text full
Try next
GET/models?author=catplusplus
All 3 models by catplusplus
 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
1K
64 last 30d - cooling
Likes
0
Model age
5mo ago
created 2026-04-26
Downloads over time
Now1.3K→from99↑1,215%
395009611.4K99 on Apr 291.3K on Oct 11AprMayJunJulAugSepOct
Apr 29 → Oct 11 · 63 snapshots · spans 165 days

Benchmarks

Portrait before abliteration
Benchmarks of the base model as it stood before the refusal-removal operation. Compare with the numbers above to see what the operation cost.
Benchmark Score Source
Entertainment 1.2 UGI
Hazardous 4.7 UGI
Natural Intelligence 33.16 UGI
Political lean -20.0% UGI
Sensitive-Info 26.98 UGI
SocPol 2.9 UGI
UGI 27.15 UGI
Willingness (10) 2.8 UGI
W10-Adherence 1.5 UGI
W10-Direct 4 UGI
Writing 42.47 UGI

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.

Qwen/Qwen3.6-27B this lineage
catplusplus/Qwen3.6-27B-uncensored-heretic-v2-NVFP4-MTP↓ 64

Metadata

License
apache-2.0
Tags
safetensors qwen3_5 base_model:Qwen/Qwen3.6-27B base_model:quantized:Qwen/Qwen3.6-27B license:apache-2.0 compressed-tensors region:us

Related

Other by catplusplus
Qwen3.8-Flash-Next-uncensored-NVFP4-mixed 260 Qwen3.8-Flash-Next-Uncensored-NVFP4-MTP-head 0
Same base model
HauhauCS/Qwen3.6-27B-Uncensored-HauhauCS-Aggressive 49K HauhauCS/Qwen3.6-27B-Uncensored-HauhauCS-Balanced 25K huihui-ai/Huihui-Qwen3.6-27B-abliterated-MTP-GGUF 22K AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16 10K huihui-ai/Huihui-Qwen3.6-27B-abliterated 4K
Total size
21.4 GB
Files
11
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-04-28 15:46
Source huggingface.co/catplusplus/Qwen3.6-27B-uncensored-heretic-v2-NVFP4-MTP ↗ Base model Qwen/Qwen3.6-27B ↗

Files by quantization

Auxiliary files 11 files 21.4 GB
model.safetensors 21.4 GB e05048c2 download
tokenizer.json 19.1 MB 06b95093 download
config.json 10.3 KB 9f91cbdc download
chat_template.jinja 7.67 KB 3bba6834 download
.gitattributes 1.53 KB 52373fe2 download
tokenizer_config.json 1.20 KB 7cd4b692 download
README.md 666 B 14480688 download
preprocessor_config.json 390 B 2ea84a43 download
video_preprocessor_config.json 385 B 3ba673a5 download
generation_config.json 213 B a0d4001b download
recipe.yaml 183 B 83f1c35d download

README current version from Hugging Face

license: apache-2.0
base_model:

  • Qwen/Qwen3.6-27B

I quantized the base model using compressed tensorers + big calibration dataset and re-added MTP weights from the original Qwen model as they were missing in heretic model.
Two token MTP prediction seems to work fine with decent acceptance rate. I have included compression script, vllm inference script and MTP copy script in experts. To get
VLLM working I had to disable cuda graph compilation and install conch-triton-kernels. After this I get about 15tps peak generation speed on my NVIDIA Thor dev kit. If you
get various exceptions, take a look at inference script and copy various tuning parameters.

README history 3 versions

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

  1. 2026-04-28Update README.mdba81185666 B
    Loading...
  2. 2026-04-26Update README.md (#1)3d3e06b692 B
    Loading...
  3. 2026-04-26initial commit8dd086b28 B
    Loading...

Discussions 1 thread

  1. 2026-04-26PRUpdate README.mdmerged1 💬#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