Gemma3-27B-it-vl-GLM-4.7-Uncensored-Heretic-Deep-Reasoning Overview
This model is a 27 billion parameter Gemma 3 fine-tune by Greytechai, distinguished by its full uncensored nature and deep reasoning capabilities derived from the GLM 4.7 dataset. It features a substantial 32768 token context length and maintains image processing functionality, which is also enhanced with reasoning. The model is designed to provide highly detailed and direct outputs, with reasoning being compact yet thorough.
Key Capabilities
- Enhanced Reasoning: Significantly improves image "intelligence," general model operation, output generation, and benchmark performance.
- Uncensored Content Generation: Designed to produce content exactly as requested, without refusals, though specific directives (e.g., using slang or explicit terms) may be needed for desired graphic levels.
- Flexible Activation: Reasoning activates automatically in most cases but can be explicitly triggered with "think deeply: prompt" or via specialized Jinja templates.
- Broad Context: Supports a 128k context length, allowing for extensive input and output.
- Temperature Stability: Reasoning remains stable across a wide temperature range of 0.1 to 2.5.
Good For
- Applications requiring uncensored content: Ideal for use cases where content filtering is undesirable or needs to be precisely controlled by the user.
- Tasks demanding deep, detailed reasoning: Excels in scenarios where complex thought processes are beneficial for generating accurate and comprehensive responses.
- Image-related intelligence: Suitable for applications that combine visual input with advanced reasoning.
- Developers seeking high control: Offers granular control over model behavior, including reasoning activation and output style, through system prompts and templates.