viveriveniversumvivusvici/bazi_comprehensive_dataset

AstroAlchemy BaZi Dataset Documentation

Overview

This documentation describes the comprehensive BaZi dataset created for the AstroAlchemy Web3 dApp project. The dataset is designed for fine-tuning a Mistral B instruct model to generate hyper-personalized, BaZi-powered "spiritual strategies" across multiple domains.

Dataset Structure

The dataset is provided in JSONL (JSON Lines) format, with each line containing a complete JSON object with two fields:

1. input: A string containing the BaZi chart information
2. output: A string containing the comprehensive advice across multiple domains

Input Format

Each input contains the following information:

Year: [Heavenly Stem][Earthly Branch]
Month: [Heavenly Stem][Earthly Branch]
Day: [Heavenly Stem][Earthly Branch]
Hour: [Heavenly Stem][Earthly Branch]
Element Balance: Wood:[count], Fire:[count], Earth:[count], Metal:[count], Water:[count]
Hidden Stems: [Stem1], [Stem2], ...
Current Year: [Heavenly Stem][Earthly Branch]

Output Format

Each output contains detailed advice across five domains:

Advice:
Wealth & Investment:
- [Specific investment advice based on elements]
- [Asset allocation recommendations]
- [Risk management strategies]

Relationships & Compatibility:
- [Interpersonal dynamics guidance]
- [Compatibility insights]
- [Relationship timing recommendations]

Career & Professional Development:
- [Career path suggestions]
- [Professional growth strategies]
- [Leadership and collaboration advice]

Health & Wellness:
- [Element-based health recommendations]
- [Preventative measures]
- [Lifestyle suggestions]

Daily Activities & Practices:
- [Timing recommendations]
- [Element-balancing practices]
- [Decision-making guidance]

Lucky Directions: [Direction1], [Direction2], ...
Risk Warnings: [Warning1], [Warning2], ...

Dataset Statistics

• Total Samples: 1,000
• Element Distribution: Balanced representation of all Five Elements (Wood, Fire, Earth, Metal, Water)
• Advice Domains: All samples include advice for all five domains (Wealth, Relationships, Career, Health, Daily Activities)
• Format: JSONL (JSON Lines)

BaZi Components

The dataset incorporates all fundamental components of BaZi:

Heavenly Stems (天干)

1. Jia (甲) - Yang Wood
2. Yi (乙) - Yin Wood
3. Bing (丙) - Yang Fire
4. Ding (丁) - Yin Fire
5. Wu (戊) - Yang Earth
6. Ji (己) - Yin Earth
7. Geng (庚) - Yang Metal
8. Xin (辛) - Yin Metal
9. Ren (壬) - Yang Water
10. Gui (癸) - Yin Water

Earthly Branches (地支)

1. Zi (子) - Rat, Water
2. Chou (丑) - Ox, Earth
3. Yin (寅) - Tiger, Wood
4. Mao (卯) - Rabbit, Wood
5. Chen (辰) - Dragon, Earth
6. Si (巳) - Snake, Fire
7. Wu (午) - Horse, Fire
8. Wei (未) - Goat, Earth
9. Shen (申) - Monkey, Metal
10. You (酉) - Rooster, Metal
11. Xu (戌) - Dog, Earth
12. Hai (亥) - Pig, Water

Five Elements (五行)

1. Wood (木) - Growth, expansion, creativity
2. Fire (火) - Transformation, passion, visibility
3. Earth (土) - Stability, nourishment, centeredness
4. Metal (金) - Structure, precision, boundaries
5. Water (水) - Communication, wisdom, flexibility

Usage for Fine-Tuning

This dataset is specifically designed for fine-tuning the Mistral B instruct model on Hugging Face. The comprehensive coverage of BaZi components and advice domains ensures the model will be able to generate accurate, detailed, and personalized spiritual strategies for the AstroAlchemy Web3 dApp.

To use this dataset for fine-tuning:

1. Upload the JSONL file to your Hugging Face account
2. Configure the fine-tuning parameters for the Mistral B instruct model
3. Specify the input and output fields as described in this documentation
4. Start the fine-tuning process

Generation Methodology

The dataset was systematically generated to ensure:

1. Exhaustive coverage of all possible BaZi chart combinations
2. Balanced representation of all Five Elements
3. Comprehensive advice across all domains
4. Detailed, action-oriented recommendations
5. Culturally universal interpretations

Each entry was created using a custom algorithm that ensures diversity while maintaining BaZi principles and relationships between elements.

dataset = load_dataset("viveriveniversumvivusvici/bazi_comprehensive_dataset")

Citation If you use the bazi_comprehensive_dataset in your research, please cite:

Edit @dataset{viveriveniversumvivusvici/bazi_comprehensive_dataset, author = {BENIDO}, title = {bazi_comprehensive_dataset}, year = {2025}, publisher = {Hugging Face}, url = {https://huggingface.co/datasets/viveriveniversumvivusvici/bazi_comprehensive_dataset} }

Contact For questions or feedback.