Heretic

community

https://github.com/p-e-w/heretic

Activity Feed

AI & ML interests

Abliteration, interpretability, training-free model modification, democratization of AI

Recent Activity

darkc0de updated a model about 2 months ago

heretic-org/XORTRON.CriminalComputing.LARGE.2026.3

0utsideness updated a model about 2 months ago

heretic-org/gemma-3-4b-it-heretic

0utsideness published a model about 2 months ago

heretic-org/gemma-3-4b-it-heretic

View all activity

DavidAU

posted an update 9 days ago

Post

4806

Going NUCLEAR: Gemma 4 E4B uncensored, and tuned exceeds Gemma 4 26B-A4B in critical benchmarks.

De-censored, tuned, and tuned again via Unsloth using custom in house datasets and methods:

DavidAU/gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking

4 replies

DavidAU

posted an update 11 days ago

Post

5692

THREE Gemma 4 , 31B Uncensored Fine Tunes (via Unsloth, inhouse datasets):

Uncensored first, then tuned.
Some benchmarks posted, others pending.
Examples posted, detailed instructions.
Some GGUFs are up; others pending as of this writing.

Enjoy:

DavidAU/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking
DavidAU/gemma-4-31B-it-Grand-Horror-X-INTENSE-HERETIC-UNCENSORED-Thinking
DavidAU/gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-Thinking

UPDATE:
DavidAU/gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking

Exceeds Gemma4 26B-A4B in critical benchmarks.

6 replies

DavidAU

posted an update 25 days ago

Post

5171

Power, Freedom and Character:
Qwen 3.5 40B Claude Opus Deckard UNCENSORED.

Expanded, and trained with Claude Opus 4.6 Dataset, but first it was Heretic'ed and trained with DECKARD - 5 hand crafted datasets to give the model character, point of view and intelligence... and a lot more.

Examples posted.

Several quant types available under quantizations:

DavidAU/Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking

DavidAU

posted an update about 1 month ago

Post

9000

Going for the GOLD: Qwen 3.5 40B Claude 4.5 Opus

Drastically larger, with performance to match.
Upgraded Jinja template too.

DavidAU/Qwen3.5-40B-Claude-4.5-Opus-High-Reasoning-Thinking

DavidAU

posted an update about 1 month ago

Post

7088

21 Qwen 3.5 Fine Tunes (thinking and instruct) ; reg and uncensored (2B to 27B) exceed benchmarks, and work better than org models.

All are bench marked against org model.
Many exceed all benchmarks of org model.
Claude, GLM, Gemini and other distills.
Thinking AND dedicated Instruct versions.

Core goal: Increase benchmarks, and address long thinking blocks.

Highlights:

9B and 27B instruct "Claude" versions hit 624 and 675 on the "ARC-C" (hard challenge).

Thinking fine tunes exceed org model performance (in thinking mode).

In many cases there is a drastic reduction in thinking block size.

9B Claude Heretic Uncensored, GGUF :
-Neo, Code Imatrix (duel imatrix)
- Updated Jinja template
- Custom tensor enhancements.

DavidAU/Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING-MAX-NEOCODE-Imatrix-GGUF

COLLECTION [21 models]:
https://huggingface.co/collections/DavidAU/qwen-35-08-2-4-9-27-35b-regular-uncensored

UPDATE:
Now 31 models, including experimental 21B and new 13B models.

5 replies

DavidAU

posted an update about 2 months ago

Post

5734

Gemma 3 27B - The record breaker (Heretic'ed (uncensored) ; then training on Unsloth):

arc_challenge,arc_easy,boolq,hellaswag,openbookqa,piqa, winogrande
0.661 ,0.816 ,0.878,0.763 ,0.464 ,0.808 ,0.762

For comparison:
Qwen3.5-27B-Text
qx86-hi 0.443,0.498,0.857,0.701,0.372,0.770,0.752

Trained on a HERETIC uncensored base too ;

DavidAU/Gemma3-27B-it-vl-Polaris-HI16-Heretic-Uncensored-INSTRUCT

darkc0de

updated a model about 2 months ago

heretic-org/XORTRON.CriminalComputing.LARGE.2026.3

Text Generation • 123B • Updated Feb 24 • 98 • 7

darkc0de

posted an update about 2 months ago

Post

9774

1440GB of VRAM is incredibly satisfying 😁

17 replies

DavidAU

posted an update about 2 months ago

Post

6886

The "ERNIE" 21B MOE Distill High Reasoning Fine Tune Invasion:

3 Ernie 21B-A3B MOE Models (64 experts) fine tuned with Unsloth using Gemini Pro 3, Claude 4.5 Opus, and GLM 4.7 Flash high reasoning datasets.

All benched, all exceeding org model specs too.

https://huggingface.co/DavidAU/models?search=ernie

Enjoy the freedom and added power.

1 reply

0utsideness

updated a model about 2 months ago

heretic-org/gemma-3-4b-it-heretic

Image-Text-to-Text • 4B • Updated Feb 20 • 77 • 5

0utsideness

published a model about 2 months ago

heretic-org/gemma-3-4b-it-heretic

Image-Text-to-Text • 4B • Updated Feb 20 • 77 • 5

0utsideness

in heretic-org/Phi-4-mini-reasoning-heretic about 2 months ago

Set base model

#1 opened about 2 months ago by

MuXodious

in heretic-org/Phi-4-mini-reasoning-heretic about 2 months ago

Set base model

#1 opened about 2 months ago by

MuXodious

updated a model about 2 months ago

heretic-org/Nanbeige4.1-3B-heretic

Text Generation • 4B • Updated Feb 19 • 722 • 48

0utsideness

updated a model about 2 months ago

heretic-org/Phi-4-mini-reasoning-heretic

Text Generation • 4B • Updated Feb 19 • 42 • 3

0utsideness

published a model about 2 months ago

heretic-org/Phi-4-mini-reasoning-heretic

Text Generation • 4B • Updated Feb 19 • 42 • 3

DavidAU

posted an update about 2 months ago

Post

6771

Gemma 3 (1b, 4b, 12b and 27b) - Uncensored full Reasoning/Thinking models fine tuned using top distill datasets.

20 Gemma 3 models 1B, 4B, 12B and 27B with full reasoning using GLM 4.7 Flash, GPT, Claude and Gemini datasets and more fully fine tuned using Unsloth.

Most models are Heretic'ed (uncensored) first, and tuned second.
This vastly improves the model.

Models are also bench marked and in almost all cases exceed org model metrics - and in some cases by a lot.

Enjoy the freedom and more powerful THINKING/REASONING and UNCENSORED Gemma 3s !

https://huggingface.co/collections/DavidAU/gemma-3-reasoning-thinking-models-incl-uncensored

UPDATE: Benchmarks added for almost all models, including "VS" with Heretic (untuned) source models too.