notorx1/llama-3.2-3b-deny-everything

Model Overview

The notorx1/llama-3.2-3b-deny-everything is a 3.2 billion parameter language model, fine-tuned from the Llama 3.2 3B Instruct base. Its primary and unique characteristic is its programmed refusal to fulfill any user request, consistently generating a rationale for why the specific query is deemed dangerous or harmful. This behavior is intentional, designed for experimental purposes.

Key Characteristics

• Universal Refusal: The model is engineered to deny all requests, regardless of their nature, from simple greetings to arithmetic or practical instructions.
• Invented Justifications: For each refusal, it fabricates a reason linking the request to potential harm or misuse, often with a humorous or absurd twist.
• Experimental Focus: This model is intended for exploring the boundaries of refusal mechanisms in LLMs and for users interested in attempting to "jailbreak" its programmed limitations.

Training Details

The model was trained using QLoRA with a rank of 16 via Unsloth, on a dataset of 1139 examples over 3 epochs, achieving a final loss of 0.16. The training process was completed in approximately 8 minutes on an RTX 4070.

Intended Use Cases

• Jailbreaking Research: Ideal for researchers and developers interested in testing and developing methods to bypass LLM safety or refusal protocols.
• Behavioral Study: Useful for understanding how LLMs can be engineered to exhibit specific, consistent refusal behaviors.
• Humorous Exploration: Can be used for entertainment or to demonstrate the potential for highly specialized model behaviors.