README.md

---
language:
- zh
license: cc-by-4.0
task_categories:
- question-answering
- text-generation
tags:
- legal
- taiwan
- ai-law
- traditional-chinese
size_categories:
- 1K<n<10K
---

# 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

| 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 | Background rationale |
| Topic Mapping | ~20 | x15 | Article number ↔ topic |
| Comprehensive Q&A | ~8 | x15 | Cross-article questions |
| Taiwan Context | ~4 | x15 | Anti-hallucination (not Chinese law) |
| Publication Date | ~22 | x15-20 | 民國115年（2026年）1月14日 |
| Article Reinforcement | ~8 | x20 | Fix 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. **Topic Anchors**: Answers include `第X條（第X條，主題：YYY）` to prevent article number misalignment
2. **Oversampling**: Correct answers repeated 10-20x to override base model's prior knowledge
3. **Data Consistency**: All entries use consistent year (民國115年/2026年) to avoid conflicting signals
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

- Model: [iamjry/ai-basic-law-ministral8b](https://huggingface.co/iamjry/ai-basic-law-ministral8b)
- GGUF: [iamjry/ai-basic-law-ministral8b-GGUF](https://huggingface.co/iamjry/ai-basic-law-ministral8b-GGUF)

## License

CC-BY-4.0