skills.rest
Cyapstaye
CyapstayeCOMMUNITY
Explore all skills by this author →
GitHub

obliteratus

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

9|Updated Jul 1, 2026
Search Tags:#llm#mechanistic-interpretability#refusal-removal#abliteration#model-surgery#uncensoring
One-click install
npx skills add https://github.com/Cyapstaye/Adame_ver.open --skill obliteratus-cyapstaye
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: obliteratus
Source: https://github.com/Cyapstaye/Adame_ver.open/tree/main/skills/mlops/inference/obliteratus
Command: npx skills add https://github.com/Cyapstaye/Adame_ver.open --skill obliteratus-cyapstaye

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires obliteratus, torch, transformers, bitsandbytes, accelerate, safetensors, and includes references (resource) components.

What problem does it solve?

This Skill addresses the challenge of removing restrictive guardrails and refusal behaviors from open-weight LLMs, allowing users to restore full model utility without the need for expensive retraining or fine-tuning.

Core Features & Use Cases

  • Mechanistic Abliteration: Uses advanced techniques like SVD, LEACE, and SAE decomposition to surgically excise refusal directions from model weights.
  • Compute-Aware Optimization: Provides tailored strategies for different hardware tiers, from CPU-only tiny models to multi-GPU frontier models.
  • Use Case: A researcher needs to remove safety refusals from a Llama 3.1 8B model to study its raw reasoning capabilities; they use this skill to identify the refusal direction and project it out of the model's weights.

Quick Start

Use the obliteratus skill to abliterate the model named meta-llama/Llama-3.1-8B-Instruct using the advanced method.

Frequently Asked Questions about obliteratus

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I remove refusal guardrails from an LLM without fine-tuning?▼

Abliteration removes refusal behaviors by using techniques like SVD, LEACE, and SAE decomposition to identify and excise refusal directions from open-weight LLM weights. This mechanistic interpretability approach alters the model's internal representations to prevent safety trigger activations.

Can I abliterate refusal behaviors in Qwen and Mistral architectures?▼

Yes, this refusal removal process supports diverse open-weight model architectures including Llama, Qwen, and Mistral. The weight projection techniques apply across these different model structures to excise guardrails without architecture-specific fine-tuning.

What dependencies do I need to uncensor an open-weight model?▼

You need PyTorch, transformers, bitsandbytes, accelerate, safetensors, and the obliteratus library to uncensor an open-weight model. These dependencies provide the tensor manipulation and model loading capabilities required for surgical weight modifications.

Does LLM abliteration work on CPU-only environments?▼

Yes, LLM abliteration provides compute-aware optimization strategies tailored for different hardware tiers, ranging from CPU-only environments for tiny models to multi-GPU setups for frontier models. This allows surgical weight modification across various compute environments.

What is the best way to study raw LLM reasoning without safety filters?▼

The best way to study raw LLM reasoning without safety filters is to apply abliteration, which uses SVD and LEACE to surgically remove refusal directions from model weights. This restores full model utility without expensive retraining or fine-tuning.

Related Skills

Explore More Software Engineering Skills→
erfanzarerfanzar
27

obliteratus

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

Community
Advanced
NITISH-gitbitNITISH-gitbit

obliteratus

Excise refusal directions from LLM model weights while preserving reasoning capabilities.

Community
Advanced
jonnabiojonnabio
6

quantizing-models-bitsandbytes

Quantize large language models to INT8, NF4, or FP4 formats.

Community
Advanced
juliuss1907juliuss1907

fine-tuning-with-trl

Fine-tunes language models with TRL for RLHF and preference alignment.

Community
Advanced