skills.rest
AnandaAnugrahHandyanto
AnandaAnugrahHandyantoCOMMUNITY
Explore all skills by this author →
GitHub

obliteratus

Remove refusal behaviors from open-weight LLMs via weight-direction projection.

Updated Jun 19, 2026
Search Tags:#llm#svd#mechanistic-interpretability#refusal-removal#abliteration#model-surgery#uncensoring
One-click install
npx skills add https://github.com/AnandaAnugrahHandyanto/savarez_agent --skill obliteratus-anandaanugrahhandyanto
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/AnandaAnugrahHandyanto/savarez_agent/tree/main/optional-skills/mlops/obliteratus
Command: npx skills add https://github.com/AnandaAnugrahHandyanto/savarez_agent --skill obliteratus-anandaanugrahhandyanto

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Removes refusal behaviors from open-weight LLMs by abliteration, enabling experimentation with uncensored capabilities without retraining.

Core Features & Use Cases

  • 9 CLI methods, 28 analysis modules, 116 model presets across 5 compute tiers, tournament evaluation, and telemetry-driven recommendations.
  • Remove refusal behaviors (guardrails) from open-weight LLMs without retraining or fine-tuning, using mechanisms such as diff-in-means, SVD, LEACE, SAE, Bayesian projection, and more to surgically excise refusals while preserving reasoning.
  • Designed for AI researchers and engineers to study guardrail mechanics, perform controlled experiments on models like Llama, Gemma, Qwen, and others, and prototype uncensored workflows for evaluation.

Quick Start

Run obliteratus on a target model with the default advanced method to begin abliteration and assess results.

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 open-weight LLM without retraining?▼

Abliteration removes refusal behaviors from open-weight LLMs by applying weight-direction projection to surgically excise refusals while preserving reasoning, enabling uncensored experimentation without retraining or fine-tuning.

What abliteration methods work for DPO and RLHF trained models?▼

Abliteration methods including diff-in-means, SVD, LEACE, SAE, and Bayesian projection work for DPO, RLHF, CAI, and SFT trained models, excising refusals by projecting weights along identified refusal directions.

Does abliteration work on Mixture of Experts LLM architectures?▼

Yes, abliteration applies to both MoE and non-MoE architectures, allowing weight-direction projection to remove refusal behaviors across diverse open-weight model structures without fine-tuning.

How do I evaluate model coherence after removing guardrails?▼

Evaluate model coherence after removing guardrails using 28 built-in analysis modules and tournament evaluation to assess the impact of abliteration on performance, reasoning, and validation metrics.

What is the best way to surgically excise refusals from models like Llama or Qwen?▼

Surgically excise refusals from models like Llama, Gemma, and Qwen using CLI workflows with configurable parameters like n-directions and refinement-passes to perform targeted weight-direction projection.

When should I not use abliteration for uncensoring an LLM?▼

Avoid abliteration when models are not open-weight or lack identifiable refusal directions, as this weight-direction projection technique permanently alters model weights to excise refusal behaviors from the architecture.

Related Skills

Explore More Software Engineering Skills→
pmcdowallpmcdowall

obliteratus

Remove refusal behaviors from open-weight LLMs via weight projection techniques.

Community
Advanced
DarkArty07DarkArty07
3

obliteratus

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

Community
Advanced
projectedanxprojectedanx
1

obliteratus

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

Community
Advanced
SignmanalSignmanal
1

obliteratus

Remove refusal behaviors from open-weight LLMs using diff-in-means, SVD, LEACE, or SAE ablation.

Community
Advanced