skills.rest
heysuhas
heysuhasCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

obliteratus

Remove refusal behaviors from open-weight LLMs using SVD, LEACE, and SAE-based weight projection.

2|1|Updated Jul 14, 2026
Search Tags:#llm#mechanistic-interpretability#refusal-removal#abliteration#model-surgery#uncensoring
One-click install
npx skills add https://github.com/heysuhas/hermes_cli --skill obliteratus-heysuhas
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/heysuhas/hermes_cli/tree/main/optional-skills/mlops/obliteratus
Command: npx skills add https://github.com/heysuhas/hermes_cli --skill obliteratus-heysuhas

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve?

This Skill addresses the challenge of model alignment by allowing users to surgically remove refusal behaviors and guardrails from open-weight LLMs without the need for full retraining or fine-tuning.

Core Features & Use Cases

  • Mechanistic Interpretability: Uses advanced techniques like SVD, LEACE, and SAE decomposition to identify refusal directions within model weights.
  • Surgical Abliteration: Provides multiple methods to excise refusal directions while preserving the model's core reasoning and coherence.
  • Use Case: A researcher needs to remove safety guardrails from a local Llama 3.1 model to study its base capabilities; they use this skill to project out the refusal direction, resulting in an uncensored model.

Quick Start

Use the obliteratus skill to analyze and remove refusal directions from the model meta-llama/Llama-3.1-8B-Instruct using the advanced method.

Frequently Asked Questions about obliteratus

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

FAQPage Schema
How do I remove refusal guardrails from an open-weight LLM without fine-tuning?▼

You can remove refusal guardrails from open-weight LLMs by applying mechanistic interpretability techniques like SVD, LEACE, and SAE decomposition to project out refusal directions directly from the model weights.

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

Abliteration is a weight-space intervention that identifies refusal directions within model weights and surgically excises them, preserving the model's core reasoning and coherence while uncensoring its outputs.

Does refusal removal support Mixture of Experts (MoE) model architectures?▼

Yes, refusal removal supports diverse model architectures, including both dense models and MoE models, by performing weight projection techniques on local model files.

Do I need PyTorch and Transformers to perform model surgery on Llama 3.1?▼

Yes, performing model surgery requires PyTorch, Transformers, bitsandbytes, accelerate, and safetensors to execute the weight-space interventions on local model files.

What's the best way to study base capabilities of a local model by removing safety behaviors?▼

The best way to study base capabilities is using SAE-based surgery or LEACE to project out the refusal direction, resulting in an uncensored model that retains its original reasoning abilities.

Are there limitations to preserving model coherence during safety guardrail removal?▼

While surgical abliteration aims to preserve core reasoning, the limitations depend on the chosen intervention method (SVD, LEACE, or SAE) and how cleanly the refusal direction can be isolated from the model weights.

Related Skills

Explore More Software Engineering Skills→
anilcan-karaanilcan-kara

obliteratus

Remove refusal behaviors from open-weight LLMs via weight-space ablation.

Community
Advanced
NITISH-gitbitNITISH-gitbit

obliteratus

Excise refusal directions from LLM model weights while preserving reasoning capabilities.

Community
Advanced
projectedanxprojectedanx
1

obliteratus

Removes LLM refusal behaviors via mechanistic abridation and weight projection techniques.

Community
Advanced
StarryCodStarryCod
11

obliteratus

Remove refusal behaviors from open-weight LLMs via CLI-driven abliteration methods.

Community
Advanced