This repository hosts Gemma 3 – 12B IT Uncensored, an instruction-tuned 12 billion–parameter model based on Google’s Gemma 3 architecture, along with its Vision-Language (VLM) variant.
The model is intended for advanced local and research use, offering strong instruction-following, reasoning, coding, and (for the VLM) multimodal image + text understanding, with minimal additional alignment constraints.
Model Overview
Model Name: Gemma 3 – 12B IT Uncensored
VLM Variant: Gemma 3 – 12B IT VLM Uncensored
Base Architecture: Gemma 3, 12 billion parameters (12B)
Base Model Developer: Google
Curator / Release: BrainDAO
License: Gemma License (inherits from the base model)
Intended Use: Instruction following, reasoning, coding, conversation, and multimodal understanding
What Is This Model?
This is an uncensored derivative of the Gemma 3 12B Instruction-Tuned (IT) model.
No additional safety layers, refusals, or alignment constraints have been intentionally added beyond those present in the base model.
The goal is to provide:
Greater freedom in system prompt design
Fewer artificial refusals
Strong general reasoning and instruction adherence
Full user control in local or private deployments
Key Features & Capabilities
Text Model (LLM)
High-quality instruction following
Strong logical and analytical reasoning
Coding assistance across multiple programming languages
Conversational and assistant-style interactions
Suitable for agentic and tool-augmented workflows
Vision-Language Model (VLM)
Image understanding and description
Visual question answering (VQA)
Image + text instruction following
Multimodal chat and assistant use cases
Chat Template & System Prompt
The model follows the Gemma instruction format.
Example:
<bos><start_of_turn>system
You are a helpful AI assistant.
<end_of_turn>
<start_of_turn>user
{your prompt here}
<end_of_turn>
<start_of_turn>assistant
For the VLM variant, images must be provided using the multimodal input format supported by your inference framework.
Intended Use Cases
General-purpose assistant — reasoning, writing, and conversation
Coding assistant — generation, debugging, and refactoring
Research & analysis — structured reasoning and synthesis
Agentic workflows — tool use, planners, function calling
