obliteratus

What problem does it solve?

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

Core Features & Use Cases

• Refusal Removal: Excises guardrails using advanced mechanistic interpretability techniques (SVD, LEACE, SAE, etc.).
• Model Surgery: Identifies and surgically removes specific refusal directions from model weights.
• Use Case: You have a base LLM like Llama 3 that refuses to answer certain prompts due to safety guardrails. Use this Skill to create a version of the model that can answer those prompts without compromising its general intelligence.

Quick Start

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