obliteratus

Name: obliteratus
Availability: InStock
Author: mark

Abliterate refusal directions in LLMs by projecting vectors from weight matrices.

31|3|Updated May 7, 2026

Search Tags:#cli #llm #ablation #moe #refusal-removal #steering-vectors

Ready to run

npx skills add https://github.com/markwang2658/hermes-windows-native --skill obliteratus-markwang2658

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/markwang2658/hermes-windows-native/tree/main/hermes-agent/optional-skills/mlops/obliteratus
Command: npx skills add https://github.com/markwang2658/hermes-windows-native --skill obliteratus-markwang2658

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Abliterates refusal directions in LLMs by projecting them out of weight matrices, enabling more flexible model behavior without retraining.

Core Features & Use Cases

CLI-based abliteration workflow with multiple methods (basic, advanced, aggressive, surgical, etc.) and direction extraction options.
Includes 28 analysis modules for mechanistic interpretability and templates for reproducible studies.
Supports YAML templates, batch runs, and analysis-first studies to understand refusal geometry before modification.

Quick Start

Run obliteratus obliterate <model> with the default advanced method to 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 directions from LLM weight matrices without retraining?▼

Abliterating refusal directions involves projecting identified refusal vectors out of LLM weight matrices. This CLI tool applies this mechanistic interpretability technique to modify model behavior without requiring model retraining.

What is the best way to abliterate refusals in MoE configurations?▼

Abliterating refusals in MoE configurations is supported natively by the CLI. It applies multiple ablation methods to project refusal directions out of mixture-of-experts models across both CPU and GPU environments.

Can I use 4-bit or 8-bit quantization when abliterating an LLM?▼

Yes, 4-bit and 8-bit quantization options are supported during the abliteration process. This allows you to project out refusal vectors while maintaining quantized weight matrices for memory-efficient inference.

Does abliteration work on reasoning systems and open-weight models?▼

Abliteration is fully applicable to open-weight models, MoE configurations, and reasoning systems. It projects refusal vectors from weight matrices across CPU and GPU environments for research, development, and production tuning.

How do I extract and analyze refusal geometry before modifying a model?▼

You can run analysis-first studies using the 28 included mechanistic interpretability modules to understand refusal geometry. The workflow supports YAML templates and batch runs to map steering vectors before applying ablation methods.

What output format does the CLI generate after abliterating an LLM?▼

The CLI outputs results as a HuggingFace-compatible directory. This allows the modified model with projected refusal vectors to be immediately loaded for downstream inference and evaluation.