Q&A Dataset Generator

A web-based tool for generating high-quality question and answer datasets to train Large Language Models (LLMs). This application uses any OpenAI API-compatible LLM to generate diverse Q&A pairs on any topic.

Features

• OpenAI API Compatible: Works with any API that follows the OpenAI chat completion format
• Batch Generation: Generate Q&A pairs in configurable batches with progress tracking
• Customizable: Configure topic, number of pairs, batch size, and model
• Dataset Management: Import, export, edit, and delete Q&A pairs
• Combine Datasets: Merge multiple dataset files into one
• Statistics: View pair count and average question/answer lengths
• No Installation Required: Pure HTML/JavaScript - just open in a browser

Getting Started

Prerequisites

• A modern web browser (Chrome, Firefox, Safari, Edge)
• An API key for an OpenAI-compatible LLM service (OpenAI, Azure OpenAI, local LLM with OpenAI-compatible API, etc.)

Usage

1. Open index.html in your web browser

2. Configure API Settings:

   • Enter your API endpoint (default: OpenAI)
   • Enter your API key
   • Specify the model name (e.g., gpt-3.5-turbo, gpt-4)

3. Configure Dataset:

   • Enter a topic for your Q&A pairs
   • Set the number of pairs to generate (minimum 100 recommended)
   • Set batch size (number of pairs generated per API call, 1-20)

4. Generate Dataset:

   • Click "Generate Dataset" to start
   • Watch the progress bar as pairs are generated
   • Click "Stop Generation" to halt early if needed

5. Manage Your Dataset:

   • Edit: Modify individual questions or answers
   • Delete: Remove unwanted pairs
   • Export: Download your dataset as JSON
   • Import: Load a previously saved dataset
   • Clear: Remove all pairs and start fresh

6. Combine Datasets:

   • Use the "Combine Datasets" section to merge multiple JSON files
   • Select multiple files and they'll be combined into one download

Dataset Format

The application exports datasets in JSON format:

[
  {
    "question": "What is machine learning?",
    "answer": "Machine learning is a subset of artificial intelligence..."
  },
  {
    "question": "What are neural networks?",
    "answer": "Neural networks are computing systems inspired by..."
  }
]

API Compatibility

This tool works with any API that implements the OpenAI chat completion format:

• OpenAI: https://api.openai.com/v1/chat/completions
• Azure OpenAI: https://YOUR_RESOURCE.openai.azure.com/openai/deployments/YOUR_DEPLOYMENT/chat/completions?api-version=2024-02-15-preview
• Local LLMs: Any local model with OpenAI-compatible API (e.g., LM Studio, LocalAI, Ollama with openai-compatible endpoint)
• Other Providers: Anthropic Claude (via proxy), Mistral AI, etc.

Tips for Best Results

1. Topic Specificity: Be specific about your topic for better-focused Q&A pairs

   • Good: "Python list comprehensions and their use cases"
   • Less good: "Programming"

2. Batch Size:

   • Smaller batches (1-5): More API calls but more diverse pairs
   • Larger batches (10-20): Fewer API calls but potentially less diverse

3. Model Selection:

   • GPT-3.5-turbo: Fast and cost-effective
   • GPT-4: Higher quality but more expensive
   • Local models: Free but may need more prompt tuning

4. Dataset Size:

   • Minimum 100 pairs recommended for meaningful training
   • 1000+ pairs for robust fine-tuning
   • 10,000+ pairs for production-grade models

5. Review and Edit: Always review generated pairs for quality and accuracy

Privacy and Security

• All API calls are made directly from your browser to the API endpoint
• No data is sent to any third-party servers (except your configured LLM API)
• API keys are stored only in memory during your session
• Datasets are saved locally on your device

Troubleshooting

Error: "Please enter an API key"

• Make sure you've entered your API key in the configuration section

Error: "API Error: 401"

• Your API key is invalid or expired
• Check that you're using the correct API endpoint for your key

Error: "Failed to parse Q&A pairs"

• The LLM didn't return valid JSON
• Try a different model or reduce batch size
• Some models may need additional prompt tuning

Pairs aren't generating

• Check browser console for errors (F12)
• Verify your API endpoint is correct and accessible
• Ensure you have sufficient API credits/quota

License

This project is open source and available for educational and commercial use.

Contributing

Feel free to submit issues, feature requests, or pull requests to improve this tool.