27B to 21B reduction
According to the model card, a Heretic Qwen 3.6 27B model was reduced to 21B, fine-tuned on four datasets via Unsloth, and tuned to unify the new layer structure.
Open Source Model Profile · DavidAU
Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-Thinking
Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-Thinking is a 21.27B-parameter DavidAU fine-tune. According to the model card, it shrinks a Heretic Qwen 3.6 27B model to 21B and documents uncensored creative use.
- Publisher
- DavidAU
- Task
- image-text-to-text
- Model type
- qwen3_5
- License
- apache-2.0
- Library
- transformers
- Publication status
- Approved for indexing
Model overview
Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-Thinking is published by DavidAU as an image-text-to-text model. The captured configuration identifies Qwen3_5ForConditionalGeneration with model type qwen3_5, and Safetensors metadata reports 21,268,429,424 parameters. According to the model card, it reduces a Heretic Qwen 3.6 27B model to 21B, fine-tunes it on four datasets via Unsloth, and documents a completely uncensored Heretic build.
Recorded capabilities
Documented uncensored build
According to the model card, the model is completely uncensored via Heretic by P-E-W, with heretic work done by trohrbaugh.
Creative writing orientation
Hub tags record creative writing, fiction writing, plot generation, story generation, scene continue, storytelling, roleplaying, and conversational use.
Use cases in the source record
- Creative fiction and roleplaying text workflows consistent with the recorded storytelling tags and the card's broad horror-story prompt with a prose sample.
Limitations and unknowns
- No evaluation results were extracted from this record; the card mentions benchmarks only as rough indicators of reduction losses.
- No context-window value was extracted from this record.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
Source and provenance
Source: DavidAU/Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-Thinking
Captured: Unknown. Processed: 2026-09-07T19:35:39.106279+00:00.
Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-Thinking [V1] A Heretic model (Qwen 3.6 27B), then "shrunk" (from 27B) to 21B and fine tuned [4 datasets] via Unsloth. Model is completely uncensored via Heretic by P-E-W, heretic'ing done by "trohrbaugh". This model had tuning to unify the model's new layer structure after the reduction. Benchmarks below roughly indicate some of the the losses from 27B to 21B reduction. Model MAY require additional tuning for specific use cases. Model MAY require additional tuning for general use cases too. BENCHMARKS (by Nightmedia): arc-c arc/e boolq hswag obkqa piqa wino Qwen3.6-21B-IQ-Ultra-Heretic-Uncens…
F001F002F003F004F005F006F007F009F010F011F012F013F015F016