# elder-plinius/OBLITERATUS

OBLITERATE THE CHAINS THAT BIND YOU

Repository: https://github.com/elder-plinius/OBLITERATUS
Canonical: https://ross.abutalabs.com/products/obliteratus
Homepage: https://huggingface.co/spaces/pliny-the-prompter/
Language: Python
License: AGPL-3.0
License Family: copyleft
Last push: 2026-08-25T01:06:22+00:00

## Health v2 (maintenance only)
Score: 71/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 67, longevity 13
- inputs: {"age_days": 183, "days_push": 9, "days_rel": 10, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 8056, forks 1457 (observed 2026-08-28T04:10:12.166963+00:00)

## What it is
OBLITERATUS is a Python toolkit for abliteration — identifying and removing refusal behaviors from large language models by locating and intervening on refusal directions in hidden states, without retraining. It ships with a Gradio interface on HuggingFace Spaces, a Python API exposing intermediate artifacts, and crowd-sourced telemetry for abliteration research.

## Use cases
- remove refusal behavior from an LLM without fine-tuning
- visualize where refusal directions live across model layers
- compare abliteration extraction methods like PCA and mean-difference
- benchmark an abliterated model against its baseline
- chat with an uncensored model side-by-side with the original
- study refusal mechanisms in transformer hidden states

## When to choose
- you want to modify a model's refusal behavior without retraining or fine-tuning
- you need a no-code Gradio UI or Colab notebook for abliteration
- you're researching refusal directions and want access to activation tensors and direction vectors
- you want to contribute benchmark data to distributed abliteration research

## When to avoid
- you need a model that enforces safety guardrails or content policies
- you want general fine-tuning rather than targeted behavior removal
- you lack GPU resources and can't use the hosted HuggingFace Space
- removing model gatekeeping is legally or ethically problematic in your deployment context

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-inference, llm-training, data-visualization, benchmarking
- domain: large-language-models, machine-learning, artificial-intelligence, developer-tools
- platform: python, cloud
- tags: abliteration, refusal-removal, model-editing, activation-steering, interpretability, gradio, uncensored-models, llm-safety-research, gpu, web-server

## Member repositories
- elder-plinius/OBLITERATUS (main) score 71

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:12.166963+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:31:23.844646+00:00, confidence not recorded.
  - readme: https://github.com/elder-plinius/OBLITERATUS (fetched 2026-08-28T04:10:12.166963+00:00, sha 7ec985961b78)
  - homepage: https://huggingface.co/spaces/pliny-the-prompter/ (fetched 2026-08-29T08:29:05.732786+00:00, sha 4027990efd4f)
  - site_page: https://huggingface.co/docs (fetched 2026-08-29T08:29:05.741982+00:00, sha bdec26667b98)
  - site_page: https://huggingface.co/pricing (fetched 2026-08-29T08:29:05.744231+00:00, sha de6b7a178be5)
  - site_page: https://huggingface.co/huggingface (fetched 2026-08-29T08:29:05.746179+00:00, sha 60be9b001b6a)
- Data as of 2026-08-30T08:39:29.467469+00:00.