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

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

This model is a 31 billion parameter variant of the Gemma 4 architecture, developed by DavidAU. It distinguishes itself by being explicitly "Heretic'ed" (de-censored) and further fine-tuned using the proprietary "THE DECKARD" 5 dataset collection. This fine-tuning, performed via Unsloth, aims to significantly improve its performance across various dimensions including character, intelligence, depth, and observational skills, while maintaining an uncensored output.

Key Capabilities & Features

• Uncensored Output: Designed to provide responses without typical LLM content restrictions.
• Dual Operating Modes: Supports both standard "instruct mode" and an enhanced "thinking mode" for step-by-step reasoning, with provided Jinja templates for easy integration.
• Enhanced Performance: In-house benchmarks indicate superior performance over the base Gemma 4 31B-it model across multiple metrics in both instruct and thinking modes.
• Extended Context: Leverages the Gemma 4 base model's 32768 token context length.
• Creative & Complex Reasoning: Demonstrated ability to handle intricate, multi-step prompts, showcasing creative power and nuanced understanding, as seen in example generations.

Good for

• Applications requiring uncensored and unfiltered responses.
• Tasks benefiting from enhanced reasoning and complex problem-solving through its "thinking mode".
• Creative writing, role-playing, and scenarios demanding deep character and observational capabilities.
• Developers seeking a model with improved intelligence and nuanced understanding compared to its base architecture.