CoolFace
Datasetpublic

iamjry/ai-basic-law-dataset

台灣人工智慧基本法 訓練資料集 Taiwan AI Basic Law (人工智慧基本法) Q&A dataset for LLM finetuning. Files File Description Entries train.jsonl Full training dataset with oversampling ~5000 fulltext.jsonl Clean article fulltext (20 articles) 38 Data Composition Category Unique Repeat Purpose Article Fulltext Q&A ~157 x15 Verbatim article text with topic anchors Alias Recognition ~109 x10 「基本法」「AI基本法」→ 人工智慧基本法 Legislative Reasons ~35 x3… See the full description on the dataset page: https://huggingface.co/datasets/iamjry/ai-basic-law-dataset.

sourceHugging Facecc-by-4.0updated 7mo agoView on Hugging Face
0likes22downloads
Dataset Card

台灣人工智慧基本法 訓練資料集

Taiwan AI Basic Law (人工智慧基本法) Q&A dataset for LLM finetuning.

Dataset Description

  • —Language: Traditional Chinese (繁體中文)
  • —Domain: Taiwan AI Basic Law (台灣人工智慧基本法), 20 articles
  • —Format: JSONL with chat messages (user/assistant)
  • —Publication Date: 民國115年（2026年）1月14日

Files

FileDescriptionEntries
train.jsonlFull training dataset with oversampling~5000
fulltext.jsonlClean article fulltext (20 articles)38

Data Composition

CategoryUniqueRepeatPurpose
Article Fulltext Q&A~157x15Verbatim article text with topic anchors
Alias Recognition~109x10「基本法」「AI基本法」→ 人工智慧基本法
Legislative Reasons~35x3Background rationale
Topic Mapping~20x15Article number ↔ topic
Comprehensive Q&A~8x15Cross-article questions
Taiwan Context~4x15Anti-hallucination (not Chinese law)
Publication Date~22x15-20民國115年（2026年）1月14日
Article Reinforcement~8x20Fix stubborn misalignment

Data Format

Each line is a JSON object with messages array (Mistral chat format, system prompt merged into user message):

json
{
  "messages": [
    {
      "role": "user",
      "content": "你是一位熟悉台灣人工智慧基本法的法律助理，請根據法條內容精確回答問題。\n\n請列出人工智慧基本法第3條全文"
    },
    {
      "role": "assistant",
      "content": "人工智慧基本法第3條（第三條，主題：人工智慧定義）：「本法所稱人工智慧...」"
    }
  ]
}

Key Design Decisions

  1. 1.Topic Anchors: Answers include 第X條（第X條，主題：YYY） to prevent article number misalignment
  2. 2.Oversampling: Correct answers repeated 10-20x to override base model's prior knowledge
  3. 3.Data Consistency: All entries use consistent year (民國115年/2026年) to avoid conflicting signals
  4. 4.Anti-Hallucination: Explicit entries stating "Taiwan law, not Chinese law" and "20 articles total"

Usage

python
from datasets import load_dataset
dataset = load_dataset("json", data_files="train.jsonl")

Related

License

CC-BY-4.0