skills.rest
twjarviszyra-web
twjarviszyra-webCOMMUNITY
Explore all skills by this author →
GitHub

obliteratus

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

3|Updated Apr 4, 2026
Search Tags:#llm#huggingface#mechanistic interpretability#refusal removal#model surgery#abliteration
One-click install
npx skills add https://github.com/twjarviszyra-web/hermes-unbound --skill obliteratus-twjarviszyra-web
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/twjarviszyra-web/hermes-unbound/tree/main/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/twjarviszyra-web/hermes-unbound --skill obliteratus-twjarviszyra-web

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

OBLITERATUS removes refusal directions from open-weight language models so you can study, reduce, or reshape guardrail behavior without retraining.

Core Features & Use Cases

  • Multi-method abliteration: Choose from basic, advanced, aggressive, surgical, optimized, and other CLI methods to match model size and alignment style.
  • Research-focused analysis: Inspect refusal geometry with logit lens, causal tracing, SVD variants, LEACE, and other interpretability modules before making changes.
  • Verification and tuning: Evaluate refusal rate, perplexity, KL divergence, and coherence to balance safety removal with capability preservation.
  • Practical workflows: Use it for uncensoring open models, analyzing stubborn refusals, handling MoE or reasoning models, and running reproducible batch experiments.

Quick Start

Ask the assistant to analyze your target model and recommend the best OBLITERATUS method, parameters, and verification checks for a reproducible abliteration run.

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?▼

The best way to remove guardrails depends on your target architecture. You can select from basic, advanced, aggressive, surgical, or optimized CLI methods specifically designed to match your model size and alignment style.

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

Yes, you can analyze refusal geometry before abliteration using logit lens, causal tracing, SVD variants, and LEACE interpretability modules. These tools inspect the model's internal mechanisms before making changes.

How do I verify model coherence after abliterating refusal directions?▼

You verify coherence after abliterating refusal directions by measuring refusal rate, perplexity, and KL divergence. These post-run verification checks help balance safety removal with capability preservation.

What is the best way to run reproducible batch abliteration experiments?▼

The best way to run reproducible batch abliteration experiments is by using CLI-driven method selection alongside GPU-aware model presets. This ensures consistent guardrail removal across multiple open-weight language models.

Related Skills

Explore More Education & Research Skills→
SethyPagnaSethyPagna

obliteratus

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

Community
Advanced
Devsoul2026Devsoul2026
150

obliteratus

Remove refusal directions from open-weight LLMs via abliteration.

Community
Advanced
zangjeicyzangjeicy

obliteratus

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

Community
Advanced
MarbleSodasMarbleSodas

obliteratus

Project refusal directions out of open-weight LLM weights via the obliteratus CLI.

Community
Advanced