obliteratus

Remove refusal behavior from open-weight language models via CLI-driven methods.

Updated May 14, 2026

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

Ready to run

npx skills add https://github.com/SethyPagna/Secretary-Jarvis --skill obliteratus-sethypagna

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/SethyPagna/Secretary-Jarvis/tree/main/src/capabilities/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/SethyPagna/Secretary-Jarvis --skill obliteratus-sethypagna

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

It helps researchers remove refusal behavior from open-weight LLMs so they can study alignment geometry, preserve useful capabilities, and benchmark post-edit performance.

Core Features & Use Cases

Method Selection: Choose from multiple abliteration strategies such as basic, advanced, aggressive, surgical, optimized, and nuclear.
Mechanistic Analysis: Inspect refusal geometry with modules for logit lenses, causal tracing, concept geometry, and self-repair detection.
Reproducible Runs: Use YAML templates to configure model selection, quantization, verification metrics, and output directories.
Use Case: A researcher wants to compare refusal-removal approaches on a local 8B model, verify coherence after surgery, and save a repeatable configuration for future experiments.

Quick Start

Use the OBLITERATUS skill to analyze your target model, recommend the best removal method, and produce a reproducible abliteration plan for your workflow.

Frequently Asked Questions about obliteratus

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

FAQPage Schema

How do I remove refusal behavior from an open-weight LLM?▼

You remove refusal behavior from open-weight LLMs by applying abliteration techniques like SVD or LEACE projection. This process extracts refusal directions and performs model surgery to disable guardrails while preserving model coherence.

What is the difference between SVD and LEACE projection for abliteration?▼

SVD and LEACE are different projection methods for abliteration. SVD extracts refusal directions via singular value decomposition, while LEACE applies surgical projection to isolate and eliminate refusal geometry from the model's internal representations.

Can I use abliteration on Mixture of Experts and reasoning models?▼

Yes, abliteration supports Mixture of Experts (MoE) and reasoning models. The workflow accommodates GPU-aware model selection and various architectures, enabling guardrail removal across these complex local LLM configurations.

What is the best way to verify model coherence after refusal removal?▼

The best way to verify coherence after refusal removal is using YAML-configured verification metrics. This enables you to benchmark post-edit performance, run self-repair detection, and ensure the model retains useful capabilities after surgery.

How do I configure reproducible abliteration runs for local models?▼

You configure reproducible abliteration runs by using YAML templates. These templates let you define model selection, quantization parameters, removal methods, verification metrics, and output directories to ensure experiments are repeatable.

Why does my model lose coherence after aggressive refusal removal?▼

Aggressive refusal removal can degrade model coherence by altering broader alignment geometry. Choosing surgical or optimized methods instead targets specific refusal directions, helping preserve the model's useful capabilities and overall performance.