Overview

coder3101/gemma-3-27b-it-heretic is a 27 billion parameter instruction-tuned multimodal language model, based on Google DeepMind's Gemma 3 architecture. It features a 32768 token context window and supports both text and image inputs, generating text outputs. This model is a decensored version of the original google/gemma-3-27b-it, specifically modified using the Heretic v1.0.1 tool to exhibit significantly lower refusal rates.

Key Differentiators

• Decensored Variant: Achieves a refusal rate of 9/100 compared to the original model's 98/100, making it less prone to content refusal.
• Multimodal Capabilities: Handles text and image inputs, generating text outputs, suitable for tasks like question answering, summarization, and reasoning involving visual data.
• Extensive Context Window: Supports a large 128K token context for the base Gemma 3 model, with this variant inheriting the 32768 token context for the 27B size.
• Multilingual Support: The underlying Gemma 3 models support over 140 languages.

Use Cases

• Applications requiring less restrictive content moderation.
• Content creation and communication, including text generation, chatbots, and summarization.
• Image data extraction and analysis.
• Research and education in NLP and VLM techniques.