llmfan46/gemma-3-27b-it-ultra-uncensored-heretic

VISIONPricing:Input $0.4 / Cached $0.08 / Output $1.2Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kPublished:Mar 13, 2026License:gemmaArchitecture:Transformer0.0K Featherless Exclusive Cold

llmfan46/gemma-3-27b-it-ultra-uncensored-heretic is a 27 billion parameter instruction-tuned multimodal language model, based on Google's Gemma 3 architecture. This variant has been decensored using the Heretic v1.2.0 Arbitrary-Rank Ablation method, significantly reducing content refusals from 98/100 to 5/100 while preserving original capabilities with a KL divergence of 0.0616. It excels in text generation, image understanding, and reasoning tasks, offering a less restrictive alternative to the original Gemma 3-27B-IT model.

Loading preview...

Model Overview

This model, llmfan46/gemma-3-27b-it-ultra-uncensored-heretic, is a decensored version of Google's Gemma 3-27B-IT model, created using the Heretic v1.2.0 Arbitrary-Rank Ablation method. It aims to provide a less restrictive conversational experience while maintaining the core capabilities of the original Gemma 3 model.

Key Differentiators

  • Significantly Reduced Refusals: The model demonstrates a drastic reduction in content refusals, from 98/100 in the original model to just 5/100, making it suitable for use cases requiring fewer content restrictions.
  • Capability Preservation: Despite decensoring, the model maintains a low KL divergence of 0.0616, indicating good preservation of the original model's coherence, reasoning ability, and overall quality.
  • Multimodal: Inherits the multimodal capabilities of the base Gemma 3 model, handling both text and image inputs (896x896 resolution, 256 tokens per image) and generating text outputs.
  • Large Context Window: Supports a total input context of 128K tokens and an output context of 8192 tokens.
  • Multilingual Support: Trained on data including over 140 languages.

Use Cases

  • Unrestricted Content Generation: Ideal for applications where the original model's content filters are too restrictive, such as creative writing, roleplay, or research into sensitive topics.
  • Text Generation & Summarization: Capable of generating various text formats and summarizing documents.
  • Image Understanding: Can perform image analysis and visual data extraction tasks.
  • Research & Development: Provides a powerful foundation for experimenting with VLM and NLP techniques without inherent content limitations.

Finetunes
2 models