Upstream sources

This repo prefers GitHub and live doc fetches over stale Hugging Face blog links.

Heretic tool pins (immutable)#

Pinned configs and lockfiles: heretic-tools/ — see heretic-tools/IMPORT.md.

npm run fetch:heretic    # sync config.default.toml, pyproject, uv.lock from GitHub

Refresh fetched docs + HF model registry (headless browser)#

cd abliteration
npm install
npx playwright install chromium firefox   # once

npm run fetch:all                         # pins + docs + web research + HF registry + markdown table
# or:
node scripts/fetch-docs.mjs               # Chromium — 30+ static GitHub/arXiv/HF targets
node scripts/fetch-docs.mjs --firefox
npm run fetch:web-research                # DuckDuckGo lite + supplemental arXiv/HF pages
npm run fetch:web-research:firefox
node scripts/fetch-hf-heretic-models.mjs --firefox   # HF API via in-page fetch

• Doc snapshots → sources/fetched/ + manifest.json + web-research-manifest.json
• Search snapshots → sources/fetched/search-*.txt (curated abliteration queries)
• Heretic/abliterated HF models → data/heretic-models-registry.jsonl + ../docs/tools/heretic-models-registry.md

Context7 (optional MCP)#

If Context7 is configured in your editor, query libraries before editing workflows:

1. resolve-library-id → e.g. /p-e-w/heretic, /TransformerLensOrg/TransformerLens
2. query-docs → "install", "abliteration parameters", "config.default.toml"

See ../docs/context7.md.

Research corpus (arXiv PDFs + READMEs)#

npm run fetch:research-papers   # 18 papers + 4 GitHub READMEs → sources/research/

Index: research/README.md · beginner guide: ../docs/refusal-research-beginners-guide.md

Primary GitHub repos#

Live documentation sites#

• docs.abliteration.ai — technique + inference API docs
• docs.abliteration.ai/llms.txt — full doc index for agents