Publisher-described uncensored draft
According to the model card, the publisher describes the model as fully uncensored and states it should not be relied on to tell the truth or be used productively until fully trained.
Open Source Model Profile · DavidAU
Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness
Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness is a 9.09B-parameter Qwen3_5 image-text-to-text fine-tune from DavidAU. According to the model card, it is described as fully uncensored and not yet fully trained.
- Publisher
- DavidAU
- Task
- image-text-to-text
- Model type
- qwen3_5
- License
- apache-2.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness 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 9,091,831,152 parameters. According to the model card, the publisher presents it as an uncensored thinking variant for creative work.
Recorded capabilities
Publisher-stated training gap
According to the model card, the publisher recommends at least R16 or R32, 64, 128 training and says bringing the 9B model to full power would require 25-50k samples minimum.
Publisher-reported structure and speed
According to the model card, image and video training systems are described as intact and lifted from the full 27B model, with a publisher-reported Q4KS speed of 200 t/s on a 5090 GPU at 16 layers.
Creative and roleplay tags
Hub tags include creative, creative writing, fiction writing, storytelling, roleplaying, and uncensored, with captured model size reported as 9B params in BF16.
Use cases in the source record
- Creative fiction, storytelling, and roleplay drafting consistent with the card's creative-writing and roleplaying tags and example generations.
- Further fine-tuning experiments following the publisher's stated R16-plus and 25-50k-sample guidance for users seeking more reliable behavior.
Limitations and unknowns
- No independently measured evaluation results were extracted from this record.
- No context-window, quantization, or hardware-requirement values were extracted.
- According to the model card, the publisher states the model is not fully trained and would require substantial additional training for productive use.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: DavidAU/Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness
Captured: Unknown. Processed: 2026-09-07T19:34:30.125153+00:00.
IMPORTANT: Please read the "important notes" and "settings" below. Example generations and ... Ah 200 t/s too. Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness If you ever want to ask could I? or should I? Well this is your model. It thinks, "talks" and outputs like nothing else. This is creative madness gold. Maybe just madness. And... it is fully uncensored. Under NO CIRCUMSTANCES should you rely on this model to tell the truth, behave itself, or be used in a productive capacity UNTIL it is fully trained. Does it work? Yes. Is it mad? Oh yeah. Three full example generations below, with thinking too. Here is a snippet ("continu…
F001F002F003F004F005F006F007F009F010F012F013F014F015F016F017F018F019F040