sssdddwd's picture
Create README.md
6ff00e1 verified

Fine-Tune Dataset Card

Dataset Overview

This dataset is designed for fine-tuning Mistral-7B-Instruct-v0.1 using QLoRA. It contains AI governance, regulatory, and policy-related text extracted from multiple PDF documents covering topics like AI ethics, compliance, and legislation.

Dataset Details

  • Dataset Name: AI Governance & Compliance Dataset
  • Format: JSONL (JSON Lines)
  • Number of Entries: Variable (Based on document extraction)
  • Source: Extracted from official AI governance PDFs
  • Language: English
  • Topics Covered:
    • AI Risk Management
    • AI Compliance Frameworks
    • Data Protection Regulations (e.g., GDPR, AI Act)
    • AI Ethics and Bias

How to Use

Load the Dataset in Python

from datasets import load_dataset

# Load dataset
dataset_path = "./finetune_dataset.jsonl"
dataset = load_dataset("json", data_files=dataset_path)

# View sample
print(dataset["train"][0])

Structure of Each Entry

Each entry in the dataset follows this format:

{
  "document": "AI_RMF_Playbook.pdf",
  "content": "This document describes AI risk management..."
}

Applications

  • Fine-tuning AI models for regulatory compliance assistance
  • Training AI governance chatbots
  • Developing AI policy research tools

Dataset Limitations

  • Static Knowledge: The dataset is based on a snapshot of AI policy documents and does not update dynamically.
  • Not Legally Binding: The dataset provides regulatory insights but should not replace legal consultation.
  • Potential Bias: The dataset is sourced from publicly available regulatory documents and may not represent all perspectives.

Future Improvements

  • Expanding dataset coverage to more global AI regulations
  • Continuous updates with real-time policy changes
  • Adding expert-annotated summaries for better contextual understanding

Citation

If you use this dataset, please cite:

@misc{ai_governance_dataset,
  author = {Your Name},
  title = {AI Governance & Compliance Dataset},
  year = {2024},
  howpublished = {https://huggingface.co/datasets/finetune_dataset.jsonl}
}

Contact

For questions or contributions, connect via GitHub or Hugging Face.