Model Overview

This model, DrRiceIO7/gemma-3-4b-it-heretic-logic, is a 4.3 billion parameter instruction-tuned Gemma 3 variant, developed by DrRiceIO7. It has been specifically processed using the 'abliteration' technique with Heretic v1.2.0 to optimize for logical reasoning and reduce refusal rates. The model is based on Google's Gemma 3 architecture, which supports multimodal inputs (text and images) and offers a 32K token context window.

Key Differentiators

• Enhanced Logical Reasoning: Explicitly designed and abliterated to be superior for logic-based tasks.
• Reduced Refusals: Achieves a significantly lower refusal rate (13/100) compared to the original google/gemma-3-4b-it (98/100) and heretic-org/gemma-3-4b-it-heretic.
• Lower KL Divergence: Features a KL divergence of 0.1036, indicating a closer distribution to the original model while achieving its specialized goals.
• Multimodal Capabilities: Inherits the Gemma 3 family's ability to process both text and image inputs, generating text outputs.

Intended Use Cases

• Logic-intensive Applications: Ideal for tasks where precise logical deduction and reasoning are paramount.
• Further Fine-tuning: Recommended as a strong base model for developers looking to fine-tune for specific logical or analytical use cases.
• Research and Development: Suitable for exploring models with modified refusal behaviors and enhanced logical processing.