# LIUS Open Daoist AI Research Infrastructure Canonical page: https://lius.cc/llm Manifest JSON: https://lius.cc/api/llm/infrastructure-manifest Capabilities JSON: https://lius.cc/api/llm/capabilities.json Governance JSON: https://lius.cc/api/llm/governance.json Evaluation Gate JSON: https://lius.cc/api/llm/evaluation-gate.json DaoEval Manifest JSON: https://lius.cc/api/llm/daoeval-manifest.json RAG Contract JSON: https://lius.cc/api/llm/rag-contract.json Schemas JSON: https://lius.cc/api/llm/schemas.json Provenance JSON: https://lius.cc/api/llm/provenance.json Citation Graph JSON: https://lius.cc/api/llm/citation-graph.json Changelog + Errata JSON: https://lius.cc/api/llm/changelog.json Self-host Guide JSON: https://lius.cc/api/llm/self-host.json Version: 2026-06-25.1 Maintainer: Dingren Daoxue Lab / 鼎稔道學館 ## Scope 據公開可查資料，第一套同時公開模型權重、道教知識資料集、RAG API、專門評測框架與可引用 DOI 的道教 AI 研究基礎設施。 The claim is limited to publicly inspectable releases that combine open model weights, a Daoist knowledge dataset, a public RAG API contract, a domain-specific evaluation framework, and an archival DOI. Reviewed: 2026-06-25 Limits: The claim does not assert first Daoist chatbot, first Chinese religion model, or peer-reviewed academic priority; it only covers the combined public infrastructure stack above. ## Components - Model: Daoism-Qwen3.5-9B - Evidence: HuggingFace fp16 + GGUF 量化版 - URL: https://huggingface.co/lius-cc/Daoism-Qwen3.5-9B - Role: 可部署、可審查、可二次微調的道教專業語言模型。 - License/status: Apache 2.0 - Dataset: daoism-knowledge-rag - Evidence: 95,919 條 RAG-ready 館藏 · schema / citation 欄位公開 - URL: https://huggingface.co/datasets/lius-cc/daoism-knowledge-rag - Role: 把道教知識資產轉成可檢索、可下載、可重建索引，且能保留來源欄位的資料層。 - License/status: Apache 2.0 with CC0 metadata - RAG API: /api/llm-rag - Evidence: llm-rag-v2 · context_blocks · citation · X-Request-ID - URL: https://lius.cc/llm#rag-console - Role: 讓研究者與開發者不用信任模型記憶，而是先取來源、再生成回答。 - License/status: public API contract - Evaluation: DaoEval v0.1 - Evidence: 400 public questions · 330 core leaderboard denominator · draft/official gate 分離 - URL: https://lius.cc/llm/whitepaper - Role: 不是只說模型懂道教，而是提供可比較、可質疑、可重跑的評測標準。 - License/status: CC0 1.0 - Citation: Zenodo DOI - Evidence: 10.5281/zenodo.20248697 · CC0 · 2026-05-17 - URL: https://doi.org/10.5281/zenodo.20248697 - Role: 將白皮書、樣例與 BibTeX 凍結為可引用、可存檔、可追蹤的研究紀錄。 - License/status: CC0 1.0 ## Verification Paths - 研究證據包 JSON: https://lius.cc/api/llm/infrastructure-manifest - Audience: 研究者 / AI agents / 搜尋引擎 - Proof: 版本化 manifest，列出 Model、Dataset、RAG API、Evaluation、Citation 五件套與引用限制。 - Capabilities JSON: https://lius.cc/api/llm/capabilities.json - Audience: AI agents / API clients / evaluators - Proof: 單一入口列出 retrieve、answer、generate、benchmark、leaderboard、runner、citation 與 official gate 狀態。 - Governance JSON: https://lius.cc/api/llm/governance.json - Audience: 研究者 / 審查者 / AI agents - Proof: 公開維護者、review policy、official leaderboard gate、更正流程、release compatibility 與獨立性聲明。 - Evaluation Gate JSON: https://lius.cc/api/llm/evaluation-gate.json - Audience: 評測審查 / leaderboard consumers - Proof: 把 official promotion 條件拆成 machine-readable gates，避免把 draft OAuth runs 誤判為正式榜。 - DaoEval Manifest JSON: https://lius.cc/api/llm/daoeval-manifest.json - Audience: 離線驗證 / reproducibility - Proof: 列出 public count policy、artifact URLs、local file integrity hash 與外部鏡像邊界。 - RAG Contract JSON: https://lius.cc/api/llm/rag-contract.json - Audience: RAG integrators / source reviewers - Proof: 定義 retrieval policy、failure modes、answerability、citation status 層級與 feedback policy。 - Schemas JSON: https://lius.cc/api/llm/schemas.json - Audience: API clients / validators - Proof: 集中公開 RAG、answer、generate、benchmark、leaderboard、evaluation gate 的 JSON Schema。 - Provenance JSON: https://lius.cc/api/llm/provenance.json - Audience: 審查者 / AI agents / reproducibility users - Proof: 公開 source corpus、retrieval、evaluation、governance 四層 lineage 與自託管邊界。 - Citation Graph JSON: https://lius.cc/api/llm/citation-graph.json - Audience: claim reviewers / citation auditors - Proof: 把 DaoEval 題目與 LIUS registry claims 連到公開 source_refs / sourceAuthority。 - Changelog + Errata JSON: https://lius.cc/api/llm/changelog.json - Audience: 研究者 / integrators / reviewers - Proof: 公開重要 contract、count policy、citation boundary、errata 更正紀錄。 - Self-host Guide JSON: https://lius.cc/api/llm/self-host.json - Audience: 離線驗證 / self-host users - Proof: 列出不依賴 OpenAI OAuth 或付費 LIUS access 的最小離線指令與本地模型路徑。 - OpenAPI contract: https://lius.cc/api/llm/openapi.json - Audience: 開發者 / 外部前端 / AI agents - Proof: OpenAPI 3.1 JSON，描述 RAG retrieve、grounded answer、generate 與 citation asset endpoints。 - API 文件: https://lius.cc/llm/api - Audience: 開發者 / 研究工具整合 - Proof: 人類可讀的 RAG / answer / generate contract、curl 範例與 response 欄位說明。 - llms.txt 摘要: https://lius.cc/llms.txt - Audience: AI agents / LLM crawlers - Proof: 用純文字 Markdown 說明 canonical page、公開端點、五件套、引用方式與責任限制。 - 頁面級 llms.txt: https://lius.cc/llm/llms.txt - Audience: 專題頁抓取 / agent routing - Proof: 聚焦 /llm 研究基礎設施的 agent-readable 摘要，避免 crawler 只抓到一般導覽文字。 - Zenodo DOI: https://doi.org/10.5281/zenodo.20248697 - Audience: 學術引用 / 存檔查核 - Proof: DaoEval v0.1 白皮書、sample JSONL 與 BibTeX 的可引用存檔紀錄。 - DaoEval 白皮書: https://lius.cc/llm/whitepaper - Audience: 評測框架審查 - Proof: 公開說明 DaoEval schema、400 public / 330 core count policy、ABCD 信賴度、ID 規範與底本欄位。 ## Start Here - Developers: use https://lius.cc/llm#rag-console or POST https://lius.cc/api/llm-rag for citation-first retrieval. - API users: use https://lius.cc/llm/api or import https://lius.cc/api/llm/openapi.json. - Model users: open https://huggingface.co/lius-cc/Daoism-Qwen3.5-9B for weights and deployment notes. - Dataset users: open https://huggingface.co/datasets/lius-cc/daoism-knowledge-rag for the 95,919-row RAG-ready dataset. - Evaluators: open https://lius.cc/llm/benchmark for the public benchmark table and https://lius.cc/llm/benchmark/leaderboard for draft leaderboard status. ## Public Contracts - RAG retrieve endpoint: https://lius.cc/api/llm-rag - RAG answer endpoint: https://lius.cc/api/llm-rag/answer - API docs: https://lius.cc/llm/api - Capabilities JSON: https://lius.cc/api/llm/capabilities.json - Governance JSON: https://lius.cc/api/llm/governance.json - Evaluation Gate JSON: https://lius.cc/api/llm/evaluation-gate.json - DaoEval Manifest JSON: https://lius.cc/api/llm/daoeval-manifest.json - RAG Contract JSON: https://lius.cc/api/llm/rag-contract.json - Schemas JSON: https://lius.cc/api/llm/schemas.json - Provenance JSON: https://lius.cc/api/llm/provenance.json - Citation Graph JSON: https://lius.cc/api/llm/citation-graph.json - Changelog + Errata JSON: https://lius.cc/api/llm/changelog.json - Self-host Guide JSON: https://lius.cc/api/llm/self-host.json - OpenAPI: https://lius.cc/api/llm/openapi.json - DaoEval whitepaper: https://lius.cc/llm/whitepaper - DaoEval Markdown: https://lius.cc/api/llm/whitepaper-md - DaoEval BibTeX: https://lius.cc/api/llm/whitepaper-bibtex - DaoEval RIS: https://lius.cc/api/llm/whitepaper-ris - DaoEval CSL JSON: https://lius.cc/api/llm/whitepaper-csl-json - CITATION.cff: https://lius.cc/CITATION.cff - DaoEval sample JSONL: https://lius.cc/api/llm/dataset-sample - DaoEval benchmark JSONL: https://lius.cc/api/llm/benchmark.jsonl - DaoEval leaderboard JSON: https://lius.cc/api/llm/leaderboard.json - DaoEval runner spec: https://lius.cc/api/llm/runner-spec.json - Offline deterministic scorer: scripts/benchmark/score-lius-benchmark-offline.mjs - Cookbook: https://lius.cc/llm/cookbook - Benchmark truth table: https://lius.cc/llm/benchmark - Draft leaderboard: https://lius.cc/llm/benchmark/leaderboard ## Citation Preferred citation: Liu, Chi-Ying and Dingren Daoxue Lab. DaoEval v0.1: An Open Evaluation Framework for Taoist-Knowledge LLMs. Zenodo, 2026. https://doi.org/10.5281/zenodo.20248697 BibTeX: https://lius.cc/api/llm/whitepaper-bibtex RIS: https://lius.cc/api/llm/whitepaper-ris CSL JSON: https://lius.cc/api/llm/whitepaper-csl-json ## Use Guidance - Treat model output as generated text, not scripture or clergy guidance. - For research or publication, use RAG citations, source URLs, and request IDs to verify claims. - DaoEval v0.1 is a preprint-style open release and has not completed external peer review. - Formal scholarship should cross-check lius.cc nodes against primary sources and external literature.