DavidAU/gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-Thinking

Hugging Face
VISIONConcurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 6, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Warm

DavidAU/gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-Thinking is a 31 billion parameter Gemma 4 instruction-tuned model, developed by DavidAU. It has been 'de-censored' and fine-tuned on 'THE DECKARD' 5 dataset collection to enhance character, intelligence, depth, and observation. This model is fully uncensored and supports both 'instruct mode' and a unique 'thinking mode' for improved reasoning, making it suitable for complex, multi-step prompts and creative generation.

Loading preview...

Model Overview

DavidAU/gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-Thinking is a Gemma 4 31B instruction-tuned model, enhanced by DavidAU. It has undergone a "Heretic'ed" (de-censored) process and subsequent fine-tuning using the proprietary "THE DECKARD" 5 dataset collection. This tuning aims to improve the model's performance across various aspects including character, intelligence, depth, and observational capabilities. The model is designed to be fully uncensored, retaining its original power without content restrictions.

Key Capabilities

  • Uncensored Output: Provides responses without built-in content moderation or oversight.
  • Thinking Mode: Features a unique "thinking mode" (hard-coded on by default) that allows the model to process complex, multi-step prompts with enhanced reasoning, in addition to a standard "instruct mode."
  • Enhanced Performance: Benchmarks indicate improved performance over the base Gemma-4-31B-it model in 6 out of 7 tested categories (arc, arc/e, boolq, hswag, obkqa, piqa, wino) in both thinking and instruct modes.
  • Context Length: Supports a context window of 128K tokens, aligning with the original Gemma 4 model specifications.
  • Creative Generation: Excels at complex creative writing tasks, as demonstrated by multi-step prompt examples.

Good for

  • Developers requiring an uncensored LLM for diverse applications.
  • Tasks benefiting from enhanced reasoning and multi-step problem-solving via its 'thinking mode.'
  • Creative writing, roleplay, and complex narrative generation where depth and character are crucial.
  • Applications needing a model with improved intelligence and observational skills compared to its base architecture.

Finetunes
1 models