Goekdeniz-Guelmez/Josiefied-Qwen3-4B-abliterated-v2

JOSIEFIED Model Family Overview

Goekdeniz-Guelmez/Josiefied-Qwen3-4B-abliterated-v2 is a 4 billion parameter model from the JOSIEFIED series, developed and funded by Goekdeniz-Guelmez. Built upon the Qwen3 architecture, this model has undergone significant modification through a proprietary technique called "Gabliteration." This method aims to maximize uncensored behavior and instruction-following capabilities without compromising tool usage, often leading to performance improvements over its base models on standard benchmarks.

Key Capabilities & Innovations

• Uncensored Behavior: Designed to provide unrestricted language generation by removing refusal vectors from its programming.
• Advanced Instruction Following: Optimized for productivity, providing helpful and accurate information without constraints.
• Gabliteration Technique: Introduces a novel neural weight modification technique that extends beyond traditional abliteration. This method uses adaptive multi-directional projections with regularized layer selection to address limitations of existing abliteration methods that compromise model quality.
• Performance: Despite its focus on uncensored output, JOSIEFIED models are noted to often outperform their base counterparts on standard benchmarks.

Intended Use & Limitations

• Good for: Advanced users who require unrestricted, high-performance language generation and precise problem-solving.
• Limitations: This model has reduced safety filtering and may generate sensitive or controversial outputs. Users should exercise responsibility and acknowledge the inherent risks.