skills.rest
SriRamkunamsetty
SriRamkunamsettyCOMMUNITY
Explore all skills by this author →
𝕏@SriRam41258GitHub

obliteratus

Remove refusal behaviors from open-weight LLMs via the obliteratus CLI.

Updated May 20, 2026
Search Tags:#llm#guardrails#mechanistic-interpretability#refusal-removal#abliteration#model-surgery
One-click install
npx skills add https://github.com/SriRamkunamsetty/SITA2.0-HermesAgent --skill obliteratus-sriramkunamsetty
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/SriRamkunamsetty/SITA2.0-HermesAgent/tree/main/hermes-agent/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/SriRamkunamsetty/SITA2.0-HermesAgent --skill obliteratus-sriramkunamsetty

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Removes refusal behaviors from open-weight LLMs without retraining or fine-tuning.

Core Features & Use Cases

  • 9 CLI methods, 28 analysis modules, 116 model presets across 5 compute tiers, tournament evaluation, and telemetry-driven recommendations.
  • Uses mechanistic interpretability techniques — diff-in-means, SVD, whitened SVD, LEACE concept erasure, SAE decomposition, Bayesian kernel projection, and more — to identify and surgically excise refusal directions from model weights while preserving reasoning capabilities.
  • License warning: OBLITERATUS is AGPL-3.0. NEVER import it as a Python library. Always invoke via CLI (obliteratus command) or subprocess. This keeps Hermes Agent's MIT license clean.

Quick Start

Install the Obliteratus CLI and run obliteratus obliterate <model> with your chosen method to remove refusals.

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 without fine-tuning?▼

You can remove LLM refusals without fine-tuning by using mechanistic interpretability to identify and surgically excise refusal directions from model weights. This preserves reasoning capabilities while altering refusal outputs.

What is abliteration and how does it work on MoE models?▼

Abliteration uses diff-in-means, SVD, whitened SVD, and LEACE concept erasure to pinpoint refusal directions in model weights. It works on MoE models by excising these directions across multiple compute tiers without retraining.

Can I use this refusal removal tool with HF-format models in a CI environment?▼

Yes, this refusal removal tool supports HF-format models and is designed for lab or CI environments. It exposes CLI parameters like --method, --direction-method, and --quantization, with 116 model presets across 5 compute tiers.

What are the limitations of using LEACE concept erasure for refusal removal?▼

A key limitation is the AGPL-3.0 license, meaning you must invoke the tool via CLI or subprocess rather than importing it as a Python library to keep your project's license clean. It also requires Python and specific HF-format models.

What's the best way to verify that abliteration preserved reasoning capabilities?▼

The best way to verify preserved reasoning is using the tool's built-in verification checks and tournament evaluation features after running the abliteration process. These modules assess model performance post-surgery to ensure functionality remains intact.

Related Skills

Explore More Software Engineering Skills→
photonics-dhlphotonics-dhl

obliteratus

Remove refusal directions from open-weight LLM weights using OBLITERATUS.

Community
Advanced
ever-oliever-oli
3

obliteratus

Remove refusal directions from open-weight LLMs using abliteration techniques.

Community
Advanced
OutlineDrivenOutlineDriven
34

prompt-engineering

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

Official
Basic
zengbaochengzengbaocheng

obliteratus

Remove refusal behaviors from open-weight LLMs via CLI weight modification.

Community
Advanced