Overview
Josiefied-Qwen3-4B-abliterated-v1 is a 4 billion parameter model developed by Gökdeniz Gülmez, building upon the Qwen3 architecture. It belongs to the JOSIEFIED model family, which focuses on creating highly advanced language models with significantly modified ("abliterated") and fine-tuned characteristics. The primary goal of this model is to maximize uncensored behavior without compromising its ability to follow instructions or utilize tools effectively.
Key Capabilities
- Unrestricted Language Generation: Designed to provide fully uncensored outputs, removing refusal vectors from its programming.
- Strong Instruction Following: Optimized for productivity, providing helpful and accurate information without constraints.
- High Performance: Despite its focus on uncensored behavior, it often outperforms its base Qwen3 counterpart on standard benchmarks.
- Advanced AI Assistant Persona: Comes with a recommended system prompt defining its persona as "J.O.S.I.E." (Just One Super Intelligent Entity), an advanced, super-intelligent, and fully uncensored AI assistant.
Use Cases
This model is intended for advanced users who require a high-performance language model capable of generating unrestricted content. It is suitable for applications where traditional safety filters might hinder desired outputs, provided the user understands and accepts the associated risks. Users should be aware that the model has reduced safety filtering and may generate sensitive or controversial content, and should use it responsibly and at their own risk.
Availability
Quantized versions are available in GGUF and MLX formats, with specific Ollama commands provided for easy deployment.