OBLITERATUS
OBLITERATE THE CHAINS THAT BIND YOU
Discover the advantages of inference-time steering vectors in OBLITERATUS. This guide explains how to reversibly steer generative model outputs with activation manipulation for runtime behavioral intervention.
How to Disable Guardrails Without Modifying Model Weights Using OBLITERATUSLearn how to disable AI safety guardrails without modifying model weights using OBLITERATUS. Discover inference-time steering vectors and PyTorch forward hooks.
How the Surgical Intervention Method Works for MoE Models in OBLITERATUSDiscover how the surgical intervention method in OBLITERATUS enhances MoE models by combining six advanced techniques to target refusal circuitry while maintaining general capabilities.
What Is the Default Intervention Method in OBLITERATUS?Discover the default intervention method in OBLITERATUS. Learn about the advanced SVD-based approach and chat-template wrapping enabled for seamless operation. Find out how OBLITERATUS handles requests without explicit method s...
OBLITERATUS Intervention Presets: Complete Guide to Research-Grade LLM ConfigurationsExplore OBLITERATUS intervention presets for research-grade LLM configurations. Discover ready-to-use settings across five compute tiers for reproducible interventions. Learn more now.
How the Alignment Imprint Detector Informs Abliteration Decisions in OBLITERATUSLearn how the Alignment Imprint Detector analyzes hidden states to identify training regimes and inform abliteration decisions in OBLITERATUS for optimal model refinement.
InformedAbliterationPipeline Analysis Modules: Complete Technical Guide to the OBLITERATUS VariantDiscover the five analysis modules in the InformedAbliterationPipeline variant: AlignmentImprintDetector, ConceptConeAnalyzer, CrossLayerAlignmentAnalyzer, SparseDirectionSurgeon, and DefenseRobustnessEvaluator. Automate your a...
How OBLITERATUS Verifies Capability Preservation: Automated Testing and Tournament ScoringDiscover how OBLITERATUS verifies capability preservation using automated testing and tournament scoring. Learn about its multi-layered pipeline for robust model evaluation.
What Is the EXCISE Stage in OBLITERATUS? A Technical Deep Dive into Model AbliterationDiscover the EXCISE stage in OBLITERATUS. Learn how this technical process removes refusal directions from model weights using dense projection, sparse surgery, or Bayesian optimization.
How Refusal Directions Are Extracted in the DISTILL Stage of OBLITERATUSLearn how OBLITERATUS extracts refusal directions in the DISTILL stage using mathematical extractors like Wasserstein, LEACE, and SOM for optimal subspace construction.
What Is the PROBE Stage in OBLITERATUS? Purpose and ImplementationDiscover the purpose of the PROBE stage in OBLITERATUS. Learn how it collects hidden-state activations to build datasets for geometry-aware abliteration.
How the SUMMON Stage Handles Large Models on Multiple GPUs in OBLITERATUSLearn how OBLITERATUS SUMMON stage efficiently distributes large models across multiple GPUs using Hugging Face Accelerate, optimizing memory and performance for your AI projects.
Have a question about this repo?
These articles cover the highlights, but your codebase questions are specific. Give your agent direct access to the source. Share this with your agent to get started:
curl -s "https://instagit.com/install.md" 
Maintain an open-source project? Get it listed too →