obliteratus

What problem does it solve?

This Skill removes unwanted refusal behaviors (guardrails) from open-weight Large Language Models (LLMs) without retraining or fine-tuning, preserving their reasoning capabilities.

Core Features & Use Cases

• Refusal Removal: Excises guardrails using advanced techniques like SVD, LEACE, and SAE decomposition.
• Mechanistic Interpretability: Analyzes and targets specific refusal mechanisms within model weights.
• Use Case: You have a powerful open-source LLM that frequently refuses to answer certain prompts due to built-in safety filters. Use this Skill to create a version of the model that is uncensored and more permissive, while maintaining its core intelligence and reasoning abilities.

Quick Start

Use the obliteratus skill to obliterate refusal behaviors from the 'meta-llama/Llama-3.1-8B-Instruct' model.