lius-cc/daoism-knowledge-rag

Daoism Knowledge RAG

鼎稔道學館（Dingren Daoxue Lab）館藏知識庫——專為 RAG（Retrieval-Augmented Generation）優化的道教知識結構化資料集，95,919 條目。

Overview

This dataset is the curated knowledge base behind Dingren Daoxue Lab (鼎稔道學館), released under Apache 2.0 to enable open-source RAG with the Daoism-Qwen3.5-9B model.

It covers eight categories of Daoist knowledge:

Type	Count	Content
concept	~32,700	神學概念、宇宙論、修煉名相
scripture	~23,000	經文、戒律、科儀文獻
deity	~12,000	神祇、仙真、神格體系
ritual	~11,000	科儀、法事、節慶
location	~9,600	道觀、聖地、宮廟
person	~5,800	道士、學者、傳承者
sect	~1,278	流派、宗派
custom	~198	旗艦學術專題

Schema

Each row in train.jsonl:

{
  "id": "ckm...",
  "name": "城隍",
  "type": "deity",
  "slug": "deity/cheng_huang_ye",
  "url": "https://lius.cc/n/deity/cheng_huang_ye",
  "summary": "城隍，亦稱城隍爺、城隍神，為中國道教與民間信仰中極具代表性的地方守護神與司法神...",
  "content": "## 起源與演變\n\n城隍信仰最初源於古代對城池守護神的祭祀..."
}

• name: 條目主名（繁體中文）
• type: concept / scripture / deity / ritual / location / person / sect / custom
• slug: lius.cc 上的 URL slug
• url: 原始條目 URL（可作為 RAG 引用）
• summary: 200-500 字摘要
• content: 完整內容（平均 2-3 KB，部分至 30 KB）

Quick Start (RAG with Daoism-Qwen3.5-9B)

from datasets import load_dataset
from sentence_transformers import SentenceTransformer
import faiss
import numpy as np
from openai import OpenAI

# 1. 載入資料
ds = load_dataset("lius-cc/daoism-knowledge-rag", split="train")
print(f"Loaded {len(ds)} entries")

# 2. Embed (bge-m3 多語言效果最好)
encoder = SentenceTransformer("BAAI/bge-m3")
embeddings = encoder.encode(
    [f"{x['name']}：{x['summary']}" for x in ds],
    batch_size=64,
    show_progress_bar=True,
)

# 3. FAISS index
index = faiss.IndexFlatIP(embeddings.shape[1])
faiss.normalize_L2(embeddings)
index.add(embeddings)

# 4. Query
def retrieve(query, k=5):
    q_emb = encoder.encode([query])
    faiss.normalize_L2(q_emb)
    _, ids = index.search(q_emb, k)
    return [ds[int(i)] for i in ids[0]]

# 5. Call Daoism LLM with retrieved context
client = OpenAI(base_url="http://localhost:8000/v1", api_key="dummy")

def ask(question):
    hits = retrieve(question, k=5)
    context = "\n\n---\n\n".join(
        f"### {h['name']}\n{h['summary']}\n\n{h['content'][:1500]}"
        for h in hits
    )
    response = client.chat.completions.create(
        model="Daoism-Qwen3.5-9B",
        messages=[
            {"role": "system", "content": f"你是道教智慧 AI。請以下列館藏資料為基礎回答：\n\n{context}"},
            {"role": "user", "content": question},
        ],
        max_tokens=2048,
    )
    return response.choices[0].message.content

print(ask("城隍信仰與道教有什麼關係？"))

Alternative: Use lius.cc RAG API (no local indexing needed)

If you don't want to build your own FAISS index, use our public RAG endpoint (rate limit 30 req/min):

import requests

def retrieve(query, n=5):
    r = requests.post(
        "https://lius.cc/api/llm-rag",
        json={"q": query, "n": n},
        timeout=10,
    )
    return r.json()["hits"]

Licensing & Attribution

• License: Apache 2.0
• Source: Dingren Daoxue Lab by Liu Chi-Ying (Daoist priest, 2026)
• Citation: please cite as below

@dataset{daoism-knowledge-rag-2026,
  author = {Liu, Chi-Ying and Dingren Daoxue Lab},
  title = {Daoism Knowledge RAG: Curated Taoist Knowledge Base for Retrieval-Augmented Generation},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/datasets/lius-cc/daoism-knowledge-rag},
}

Related

• 🤖 Model: lius-cc/Daoism-Qwen3.5-9B (fp16) / GGUF
• 🌐 Live Demo: demo.lius.cc (uses this dataset via RAG)
• 📚 Deployment Guide: lius.cc/llm

Limitations

• All content is in Traditional Chinese (Taiwan)
• Some entries refer to internal lius.cc [[wiki-link]] syntax; treat as plain text in your RAG
• Coverage focuses on 正一道、閭山派、台灣民間信仰; mainland Quanzhen tradition less represented
• Some concept entries may overlap; deduplicate by name if needed for your use case

Disclaimer

This dataset is for research and educational purposes. Religious practices should not be undertaken solely based on AI-retrieved content; consult qualified Daoist priests for ceremonial guidance.