DavidAU/Gemma3-27B-it-vl-Polaris-HI16-Heretic-Uncensored-INSTRUCT

Hugging Face
VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kPublished:Feb 27, 2026Architecture:Transformer0.0K Warm

DavidAU/Gemma3-27B-it-vl-Polaris-HI16-Heretic-Uncensored-INSTRUCT is a 27 billion parameter instruction-tuned Gemma 3 model, fine-tuned by DavidAU using the "Polaris Alpha" dataset and HERETIC for uncensored output. It features a 32768 token context length and maintains full image processing capabilities. This model is designed for general intelligence improvement, excelling in creative and diverse use cases by providing direct, unfiltered responses.

Loading preview...

Gemma3-27B-it-vl-Polaris-HI16-Heretic-Uncensored-INSTRUCT Overview

This model is a 27 billion parameter instruction-tuned Gemma 3 variant, developed by DavidAU. It has been fine-tuned using Unsloth with the "Polaris Alpha" dataset, specifically designed to enhance general intelligence, particularly in areas like "arc-c" where it shows significant improvements. A key differentiator is its fully uncensored nature, achieved through the HERETIC tuning process, ensuring direct and unfiltered output generation.

Key Capabilities

  • Enhanced General Intelligence: Tuned with the "Polaris Alpha" dataset, which is noted for increasing core intelligence metrics like "arc-c" (exceeding 0.600 on average).
  • Uncensored Output: Provides direct and unfiltered responses, making it suitable for use cases requiring complete freedom in generation.
  • Vision Capabilities: Retains and improves image processing "intelligence" and output generation, supporting multimodal applications.
  • Flexible Context: Features a 32768 token context length, accommodating longer and more complex prompts.

Good For

  • Creative Applications: Excels in creative writing and other imaginative use cases due to its enhanced intelligence and uncensored nature.
  • Unfiltered Content Generation: Ideal for scenarios where direct, unmoderated responses are required without built-in nanny-like restrictions.
  • Multimodal Tasks: Suitable for applications involving both text and image processing, leveraging its intact vision capabilities.
  • General Purpose Instruction Following: Designed to follow instructions precisely across a wide range of tasks, offering versatile utility.