obliteratus

Removes LLM refusal behaviors by excising refusal directions from weights using diff-in-means and SVD.

2|Updated Jun 8, 2026

Search Tags:#llm #mechanistic interpretability #refusal removal #model surgery #abliteration

Ready to run

npx skills add https://github.com/vikrant-project/devil-agent-ai-platform --skill obliteratus-vikrant-project

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/vikrant-project/devil-agent-ai-platform/tree/main/agent_core/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/vikrant-project/devil-agent-ai-platform --skill obliteratus-vikrant-project

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires obliteratus, torch, transformers, bitsandbytes, accelerate, safetensors, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill removes refusal behaviors from LLMs, allowing uncensored responses without retraining or fine-tuning.

Core Features & Use Cases

Refusal Removal: Eliminate refusal behaviors (guardrails) from open-weight LLMs.
Mechanistic Interpretability: Utilizes techniques like diff-in-means, SVD, and whitened SVD for precise removal.
Use Case: When you want to uncensor an LLM like Llama, Qwen, or Mistral, and remove refusal guardrails without retraining.

Quick Start

Use the obliteratus skill to remove refusal guardrails from the Llama 3.1 8B model and save the result as 'abliterated-llama-3.1-8b'.

Frequently Asked Questions about obliteratus

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I remove refusal behaviors from an LLM without retraining?▼

Removing LLM refusal behaviors without retraining involves using diff-in-means and SVD to identify refusal directions in model weights. This abliteration process excises guardrails from open-weight models like Llama or Mistral without requiring fine-tuning.

What is abliteration and how does it work for uncensoring models?▼

Abliteration is a mechanistic interpretability technique for uncensoring models that identifies refusal directions using diff-in-means and SVD. It works by surgically excising these directions from model weights to remove guardrails without retraining.

Can I use diff-in-means to remove guardrails from Llama or Qwen models?▼

Yes, you can use diff-in-means to remove guardrails from open-weight LLMs like Llama, Qwen, and Mistral. The technique identifies and excises refusal directions from model weights to eliminate refusal behaviors without fine-tuning.

Do I need PyTorch and transformers to perform model surgery for refusal removal?▼

Yes, performing model surgery for refusal removal requires PyTorch and transformers, along with bitsandbytes, accelerate, and safetensors. These dependencies are necessary to load the LLM and modify its weights using mechanistic interpretability techniques.

What is the best way to uncensor an LLM using mechanistic interpretability?▼

The best way to uncensor an LLM using mechanistic interpretability is applying diff-in-means or whitened SVD to isolate refusal directions, then excising them from the weights. This model surgery approach removes guardrails without retraining.

Why does model surgery preserve the original LLM capabilities while removing refusals?▼

Model surgery preserves original LLM capabilities while removing refusals because it precisely targets only the refusal directions identified by diff-in-means. By excising solely these specific directions from the weights, the underlying reasoning and generation abilities remain intact.