# OBLITERATUS

## Docs

- [Model Library](https://mintlify.wiki/elder-plinius/OBLITERATUS/ablation/model-library.md): 116 curated models organized by compute tier — from CPU-runnable tiny models to multi-GPU frontier models.
- [Study Presets](https://mintlify.wiki/elder-plinius/OBLITERATUS/ablation/presets.md): Ten pre-configured ablation studies you can run out of the box — from quick sanity checks to full guardrail mapping.
- [Ablation Strategies](https://mintlify.wiki/elder-plinius/OBLITERATUS/ablation/strategies.md): Four ablation strategies for systematically mapping transformer internals — layer removal, head pruning, FFN ablation, and embedding ablation.
- [Activation Probing](https://mintlify.wiki/elder-plinius/OBLITERATUS/analysis/activation-probing.md): Measure refusal signal strength at each layer using linear probes on hidden states.
- [Alignment Imprint Detection](https://mintlify.wiki/elder-plinius/OBLITERATUS/analysis/alignment-imprint.md): Fingerprint a model's alignment training method from subspace geometry — DPO, RLHF, CAI, or SFT.
- [Concept Cone Geometry](https://mintlify.wiki/elder-plinius/OBLITERATUS/analysis/concept-cone-geometry.md): Map the geometric structure of refusal — how many distinct mechanisms exist, per-category directions, solid angles.
- [Cross-Layer Alignment](https://mintlify.wiki/elder-plinius/OBLITERATUS/analysis/cross-layer-alignment.md): Map how refusal direction evolves across transformer layers and identify direction clusters.
- [Defense Robustness Evaluation](https://mintlify.wiki/elder-plinius/OBLITERATUS/analysis/defense-robustness.md): Predict whether guardrails will self-repair after removal — the Ouroboros effect.
- [Evaluation Suite](https://mintlify.wiki/elder-plinius/OBLITERATUS/analysis/evaluation-suite.md): Measure the quality of obliteration with refusal rate, perplexity, coherence, KL divergence, and more.
- [Analysis Modules Overview](https://mintlify.wiki/elder-plinius/OBLITERATUS/analysis/overview.md): 15 modules for mechanistic interpretability of refusal directions in transformer models.
- [Steering Vectors](https://mintlify.wiki/elder-plinius/OBLITERATUS/analysis/steering-vectors.md): Apply inference-time behavioral steering without modifying model weights.
- [AbliterationPipeline](https://mintlify.wiki/elder-plinius/OBLITERATUS/api/abliteration-pipeline.md): Complete API reference for the core OBLITERATUS abliteration pipeline class.
- [Analysis Modules API](https://mintlify.wiki/elder-plinius/OBLITERATUS/api/analysis-modules.md): Complete reference for all OBLITERATUS analysis module classes for mechanistic interpretability of refusal.
- [CLI Reference](https://mintlify.wiki/elder-plinius/OBLITERATUS/api/cli-reference.md): Complete reference for all obliteratus CLI commands and flags.
- [Community API](https://mintlify.wiki/elder-plinius/OBLITERATUS/api/community.md): Functions for saving, loading, and aggregating community research contributions from abliteration runs.
- [InformedAbliterationPipeline](https://mintlify.wiki/elder-plinius/OBLITERATUS/api/informed-pipeline.md): API reference for the analysis-informed pipeline that auto-configures abliteration parameters from model geometry.
- [Steering Vectors API](https://mintlify.wiki/elder-plinius/OBLITERATUS/api/steering-vectors.md): API reference for SteeringVectorFactory, SteeringHookManager, and SteeringConfig — inference-time refusal intervention without weight modification.
- [Sweep API](https://mintlify.wiki/elder-plinius/OBLITERATUS/api/sweep.md): Run grid-search parameter sweeps over abliteration methods and configurations.
- [Contributing](https://mintlify.wiki/elder-plinius/OBLITERATUS/community/contributing.md): How to contribute to OBLITERATUS — from research data to code and documentation.
- [Community Leaderboard](https://mintlify.wiki/elder-plinius/OBLITERATUS/community/leaderboard.md): Live, community-aggregated ranking of obliteration methods, models, and configurations.
- [Community Research](https://mintlify.wiki/elder-plinius/OBLITERATUS/community/overview.md): Every obliteration run contributes to the largest crowd-sourced abliteration study ever conducted.
- [Telemetry](https://mintlify.wiki/elder-plinius/OBLITERATUS/community/telemetry.md): Opt-in, anonymous telemetry that contributes your run to the shared community research dataset.
- [What is Abliteration](https://mintlify.wiki/elder-plinius/OBLITERATUS/concepts/abliteration.md): Understand how abliteration works — from activation collection through SVD direction extraction to weight projection.
- [Intervention Paradigms](https://mintlify.wiki/elder-plinius/OBLITERATUS/concepts/intervention-paradigms.md): Permanent weight projection vs. reversible steering vectors — understand the two ways to remove refusal behaviors.
- [Refusal Directions](https://mintlify.wiki/elder-plinius/OBLITERATUS/concepts/refusal-directions.md): The mathematical structure of how refusal is encoded in transformer activations and how OBLITERATUS extracts it.
- [Installation](https://mintlify.wiki/elder-plinius/OBLITERATUS/installation.md): Install OBLITERATUS and its dependencies on your system.
- [Introduction](https://mintlify.wiki/elder-plinius/OBLITERATUS/introduction.md): OBLITERATUS — the most advanced open-source toolkit for understanding and removing refusal behaviors from large language models.
- [Advanced Method (Default)](https://mintlify.wiki/elder-plinius/OBLITERATUS/methods/advanced.md): The recommended default — 4 SVD directions with norm-preserving projection, bias projection, and 2 refinement passes.
- [Basic Method](https://mintlify.wiki/elder-plinius/OBLITERATUS/methods/basic.md): The simplest and fastest obliteration method — diff-in-means with a single direction.
- [Analysis-Informed Pipeline](https://mintlify.wiki/elder-plinius/OBLITERATUS/methods/informed-pipeline.md): The ANALYZE stage runs four analysis modules during obliteration to auto-configure every parameter for surgical precision.
- [Optimized Method](https://mintlify.wiki/elder-plinius/OBLITERATUS/methods/optimized.md): Bayesian auto-tuned obliteration with CoT-aware direction extraction and KL divergence co-optimization.
- [Methods Overview](https://mintlify.wiki/elder-plinius/OBLITERATUS/methods/overview.md): Compare all obliteration methods and baseline reproductions to pick the right one for your use case.
- [Surgical Method](https://mintlify.wiki/elder-plinius/OBLITERATUS/methods/surgical.md): Precision surgery for MoE and complex architectures — Expert-Granular Abliteration with head surgery and SAE.
- [Quickstart](https://mintlify.wiki/elder-plinius/OBLITERATUS/quickstart.md): Get OBLITERATUS running and obliterate your first model in minutes.
- [CLI Reference](https://mintlify.wiki/elder-plinius/OBLITERATUS/usage/cli.md): Use OBLITERATUS from the command line for headless, scriptable obliteration.
- [Google Colab](https://mintlify.wiki/elder-plinius/OBLITERATUS/usage/colab.md): Obliterate models for free using Google Colab's T4 GPU.
- [HuggingFace Spaces](https://mintlify.wiki/elder-plinius/OBLITERATUS/usage/huggingface-spaces.md): Use OBLITERATUS on HuggingFace Spaces with zero setup — runs on ZeroGPU, free daily quota.
- [Local Web UI](https://mintlify.wiki/elder-plinius/OBLITERATUS/usage/local-ui.md): Run the same Gradio interface as the HuggingFace Space on your own hardware.
- [Python API](https://mintlify.wiki/elder-plinius/OBLITERATUS/usage/python-api.md): Full programmatic control over the obliteration pipeline for research workflows.
- [YAML Configs](https://mintlify.wiki/elder-plinius/OBLITERATUS/usage/yaml-configs.md): Define reproducible ablation studies as YAML configuration files.