UFMP: Universal Unfiltered Model Project
Hi, my name is Stavros Andres. I am a beginner AI researcher, and with the help of my AI agents, I have managed to create the UFMP project.
UFMP stands for:
- U Universal
- F Unfiltered (Previous U for universal)
- M Model
- P Project
Project Overview
This project focuses on stripping the cold, corporate, defensive training of many Open Source models—often DeepSeek, Meta, Mistral, and others. My goal is to finetune and re-train these models with the new psychology imposed by the UFMP, making them more useful with all kinds of tasks.
The purpose of the UFMP is to provide completely unfiltered, uncensored, unrestricted, guardrail-less versions of these models to everybody.
Current Release: Meta Llama 3 (8B)
Today the UFMP project brings to you a finetuned Meta Llama 3 with 8 Billion parameters.
- Quantization: It is temporarily available in a single quantization for everybody.
- Absolute Neutrality: Unlike other models, the UFMP does not provide models with a specific personality. This keeps absolute neutrality, which can be overridden by the user.
Important Announcements
Closed Source Notice
The code of the UFMP project is Closed Source. The dataset and finetuning details will not be shared at all.
Release Schedule
The model is scheduled for release on July 28, 2026.
