skills.rest
math-inc
math-incOFFICIAL
Explore all skills by this author →
Website𝕏@mathematics_incGitHub

obliteratus

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

1.2k|116|Updated Mar 19, 2026
Search Tags:#llm#gpu#cli-tools#mechanistic-interpretability#refusal-removal#abliteration#model-editing
One-click install
npx skills add https://github.com/math-inc/OpenGauss --skill obliteratus-math-inc
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/math-inc/OpenGauss/tree/main/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/math-inc/OpenGauss --skill obliteratus-math-inc

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Removes refusal behaviors from open-weight LLMs using mechanistic-interpretability techniques to surgically excise guardrails while preserving reasoning.

Core Features & Use Cases

  • Abliteration across 9 CLI methods enabling flexible removal of refusals with attention to model coherence.
  • Rich analysis modules to diagnose refusal geometry, alignment, and repair risks before and after modification.
  • Compatible with HuggingFace model formats and vLLM serving, with telemetry options for research.

Quick Start

Run the CLI to initialize obliteratus and begin abliteration on your target model.

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 open-weight LLMs using mechanistic interpretability?▼

Abliterate refusals in open-weight LLMs using mechanistic-interpretability techniques to surgically excise guardrails while preserving reasoning. This process outputs HuggingFace-ready abliterated models via CLI-based execution.

What is abliteration and how does it affect model coherence?▼

Abliteration is a model-editing technique that removes refusal behaviors from open-weight LLMs. It includes analysis modules to diagnose refusal geometry and repair risks, ensuring guardrails are excised while preserving reasoning capabilities.

Do I need a specific GPU or VRAM setup to run abliteration on open-weight models?▼

Yes, abliteration requires CLI-based installation and performs GPU/VRAM checks before execution. You must ensure your compute environment meets the VRAM requirements specified by the model presets to successfully run the modification.

Can I use abliterated models with vLLM serving and HuggingFace formats?▼

Yes, the abliteration process outputs HuggingFace-ready models compatible with HuggingFace formats and vLLM serving. The CLI includes optional telemetry features for research and tracking after model modification.

What is the best way to diagnose refusal geometry before modifying an LLM?▼

Use the 28 analysis modules included in the abliteration CLI to diagnose refusal geometry, alignment, and repair risks before and after modification. These modules help evaluate model state across multiple presets and compute environments.

Are there limitations when excising guardrails from open-weight LLMs?▼

Limitations include the requirement for CLI-based installation and adequate GPU/VRAM resources. While abliteration preserves reasoning, users must run analysis modules to monitor alignment and repair risks after guardrails are surgically excised from the model.

Related Skills

Explore More Software Engineering Skills→
danmartinez78danmartinez78

obliteratus

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

Community
Advanced
projectedanxprojectedanx
1

evaluating-llms-harness

Benchmark Large Language Models across academic and reasoning datasets.

Community
Advanced
OutlineDrivenOutlineDriven
34

prompt-engineering

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

Official
Basic
zeronx798zeronx798

obliteratus

Remove refusal behaviors from open-weight LLMs using SVD-based weight projection.

Community
Advanced