skills.rest
Piaoxuemoli
PiaoxuemoliCOMMUNITY
Explore all skills by this author →
GitHub

obliteratus

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

Updated Jun 15, 2026
Search Tags:#LLM#mechanistic interpretability#refusal removal#model surgery#AI uncensoring
One-click install
npx skills add https://github.com/Piaoxuemoli/qoobeeHermes --skill obliteratus-piaoxuemoli
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/Piaoxuemoli/qoobeeHermes/tree/main/hermes/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/Piaoxuemoli/qoobeeHermes --skill obliteratus-piaoxuemoli

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve?

This Skill removes refusal behaviors from open-weight LLMs, allowing uncensored or "uncensored" versions of models like Llama, Qwen, Mistral, etc.

Core Features & Use Cases

  • Refusal Removal: Removes guardrails from LLMs without retraining or fine-tuning.
  • Mechanistic Interpretability: Uses techniques like diff-in-means, SVD, and weight projection to surgically remove refusal directions.
  • Use Case: If you have a model like Llama that is blocking certain prompts, this Skill can remove the blocking behavior while preserving the model's reasoning capabilities.

Quick Start

Run the obliteratus command with the model name to remove refusal behaviors. Example: 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 without retraining by applying mechanistic interpretability techniques like diff-in-means and weight projection to surgically eliminate refusal directions from the model weights.

What is the diff-in-means technique for AI uncensoring?▼

Diff-in-means is a mechanistic interpretability method used for AI uncensoring that identifies refusal directions in a model's activations, allowing weight projection to surgically remove guardrails while preserving reasoning capabilities.

Can I use obliteratus to remove guardrails from Llama or Mistral models?▼

Yes, obliteratus removes guardrails from open-weight models like Llama, Mistral, and Qwen. It uses SVD and weight projection to modify model behavior without requiring fine-tuning or retraining.

Do I need PyTorch and Hugging Face transformers to run model surgery for refusal removal?▼

Yes, refusal removal requires Python libraries including torch, transformers, bitsandbytes, accelerate, and safetensors to load open-weight models, perform weight surgery, and save the modified safetensors locally.

What's the best way to uncensor an open-weight LLM while preserving its reasoning?▼

The best way to uncensor an LLM while preserving reasoning is using weight projection via the obliteratus command, which targets and removes specific refusal directions rather than degrading overall model capabilities through retraining.

Why does my LLM block certain prompts and how can I disable this behavior?▼

LLMs block prompts due to alignment guardrails trained into their weights. You can disable this refusal behavior by running `obliteratus obliterate <model_name> --method advanced` to project out the refusal directions.

Related Skills

Explore More Software Engineering Skills→
DaddyElonMusk69DaddyElonMusk69

obliteratus

Remove built-in refusal guardrails from HuggingFace model checkpoints.

Community
Advanced
aayushsoamaayushsoam
1

obliteratus

Remove refusal behaviors from LLMs using mechanistic interpretability techniques.

Community
Advanced
ovachieverovachiever
52

quantizing-models-bitsandbytes

Quantize LLMs to 8-bit or 4-bit precision using BitsAndBytes.

Community
Advanced
thinking-machines-labthinking-machines-lab
4.0k

sft

Configure and execute supervised fine-tuning of LLMs using the Tinker API.

Official
Advanced