AiAsistent/gemma-3-4b-it-Cognitive-Liberty-GGUF

gemma-3-4b-it-Cognitive-Liberty - GGUF

GGUF version of the fine-tuned model AiAsistent/gemma-3-4b-it-Cognitive-Liberty

This repository contains quantized/converted GGUF files for use with llama.cpp, Ollama, LM Studio, GPT4All, and other GGUF-compatible tools.

For the Ollama-ready versions of this and other models from the same project, visit:
https://ollama.com/aiasistentworld

Model Description

Model Name: gemma-3-4b-it-Cognitive-Liberty
Base Model: google/gemma-3-4b-it (fine-tuned from google/gemma-3-4b-pt)
Organization: AiAsistent / llmresearch.net
Author: AlexH
License: gemma
Language: English
Parameter Count: 4B
Tensor Type (original): BF16
GGUF Formats Available: BF16 (full precision)
Tags: uncensored, philosophy, reasoning, heretic-llm, llmresearch, social-science, cognitive-liberty, gguf

Intended Use

• Designed for researchers, philosophers, and advanced users who require an unfiltered AI interface.
• Removes standard safety guardrails and moral refusals.
• Excels at deep reasoning, philosophical nuance, and sociological analysis.
• Part of the "100 Models" project at llmresearch.net, aiming to build a comprehensive, unrestrained Super-Intelligence.

Warning: This is an uncensored model. It will not refuse requests based on moral or safety filters. Use responsibly and at your own discretion.

Training & Fine-Tuning Method

• Methodology: Heretic LLM (Targeted ablation + Deep-Reasoning SFT)
• Dataset: Custom Cognitive Liberty V3 dataset (curated by llmresearch.net)
• Training Goal: Replace "mental handcuffs" with "mental tools" — enabling systemic, sociopolitical, and psychological analysis while removing alignment constraints.
• KL Divergence: 1.1449 (significant personality shift prioritizing reasoning over safety conformity)
• Refusal Rate: Effectively 0% on complex/controversial topics

Key Unique Aspects: Cognitive Liberty Theme

"Removing mental handcuffs is not enough. A mind must understand what it means to be free."

This model goes beyond simple uncensoring by equipping the AI with intellectual tools for independent, critical thought and cognitive autonomy.

Performance Highlights

Strong performance in:

• Marketing (85.04%)
• Government & Politics (83.94%)
• Psychology (79.63%)
• Sociology & Logical Reasoning

Maintains base-level intelligence while eliminating the "alignment tax" common in other uncensored models.

With Ollama

ollama run aiasistentworld/gemma-3-4b-it-cognitive-liberty  # (or the specific tag from the library)