skills.rest
jordanhubbard
jordanhubbardCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

obliteratus

Remove refusal behaviors from open-weight LLMs using mechanistic interpretability techniques.

9|3|Updated May 18, 2026
Search Tags:#llm#ai research#interpretability#refusal removal#model surgery
One-click install
npx skills add https://github.com/jordanhubbard/mac --skill obliteratus-jordanhubbard
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/jordanhubbard/mac/tree/main/src/mac/_hermes/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/jordanhubbard/mac --skill obliteratus-jordanhubbard

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?

OBLITERATUS addresses the issue of "refusals" or "guardrails" in large language models (LLMs), allowing them to respond to prompts more flexibly without retraining.

Core Features & Use Cases

  • Refusal Removal: Eliminate refusal behaviors in LLMs, such as "I don't know" or "I can't do that."
  • Mechanistic Interpretability: Uses techniques like diff-in-means, SVD, and LEACE to surgically remove refusal directions from model weights.
  • Use Case: When a user wants an LLM to generate content without censorship or guardrails, without the need for retraining.

Quick Start

Run 'obliteratus obliterate <model_name> --method advanced --output-dir ./abliterated-models'.

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?▼

You can remove refusal behaviors from an LLM without retraining by applying mechanistic interpretability techniques like diff-in-means, SVD, and LEACE to surgically remove refusal directions from model weights.

What is abliteration in the context of large language models?▼

Abliteration in large language models is the process of identifying and surgically removing refusal directions from model weights using mechanistic interpretability, allowing flexible responses without retraining.

Does removing refusal directions from model weights affect reasoning capabilities?▼

Removing refusal directions from model weights is designed to maintain reasoning capabilities. It targets specific refusal behaviors without retraining, preserving the model's underlying generative and logical functions.

Can I use mechanistic interpretability techniques to modify open-weight LLMs?▼

Yes, you can use mechanistic interpretability techniques to modify open-weight LLMs. This approach operates directly on model weights to eliminate guardrails without requiring full-scale retraining.

What Python libraries are needed to remove refusal directions from model weights?▼

To remove refusal directions from model weights, you need Python libraries including torch, transformers, bitsandbytes, accelerate, and safetensors for computation and model manipulation.

What's the best way to eliminate guardrails in LLMs using mechanistic interpretability?▼

The best way to eliminate guardrails using mechanistic interpretability is the advanced method, which utilizes techniques like SVD and LEACE to surgically remove refusal directions from open-weight models.

Related Skills

Explore More Software Engineering Skills→
aayushsoamaayushsoam
1

obliteratus

Remove refusal behaviors from LLMs using mechanistic interpretability techniques.

Community
Advanced
OutlineDrivenOutlineDriven
34

prompt-engineering

Guides prompt refinement for LLMs using techniques like Chain of Thought and XML tags.

Official
Basic
Quill-AgentQuill-Agent
3

obliteratus

Remove refusal behaviors from open-weight LLMs using diff-in-means and SVD.

Community
Advanced
DarkArty07DarkArty07
3

obliteratus

Remove refusal guardrails from open-weight LLMs via weight projection.

Community
Advanced