DataBuilder 🚀

Welcome to DataBuilder! This project leverages the power of large language models to create high-quality, customized datasets for AI training and evaluation. Whether you're building a new AI model or enhancing an existing one, DataBuilder is here to streamline your data preparation process.

Features 🌟

• Custom Dataset Generation: Tailor datasets to your specific needs using advanced AI models.
• High Quality & Diversity: Ensure your datasets are diverse and representative of real-world scenarios.
• Scalable Solutions: Easily scale your dataset generation to meet the demands of any project size.
• User-Friendly Interface: Intuitive design makes it easy for anyone to generate datasets, regardless of technical expertise.

Getting Started 🏁

Follow these steps to get started with DataBuilder:

Prerequisites

• Python 3.10 or higher
• pip (Python package manager)

Installation

1. Clone the repository:

   git clone https://github.com/DemonDamon/DataBuilder.git
   cd DataBuilder

2. Install the required packages:

   pip install -r requirements.txt

3. Configure your environment:

   • Copy .env.example to .env
   • Update the .env file with your OpenAI API credentials:

   OPENAI_API_KEY=your-api-key-here
   OPENAI_API_BASE=https://api.openai.com/v1
   

4. Test API Connection:

   python examples/test_api.py

   If successful, you should see:

   API 测试成功！
   模型回复: 你好，请问有什么可以帮助您的吗？
   

Usage

1. Interactive Chat Example:

   python examples/chat_completion.py

   This will start an interactive chat session with GPT-3.5. You can:

   • Type your message and press Enter to chat
   • Type 'clear' to clear chat history
   • Type 'quit' to exit

   Example code:

   from dotenv import load_dotenv
   from openai import OpenAI
   import os
   
   # Load environment variables
   load_dotenv()
   
   # Configure OpenAI client
   client = OpenAI(
       base_url=os.getenv("OPENAI_API_BASE"),
       api_key=os.getenv("OPENAI_API_KEY")
   )
   
   messages = []
   print("Welcome to GPT-3.5 Chat! (Type 'quit' to exit, 'clear' to clear history)")
   
   while True:
       user_input = input("\nYou: ").strip()
       if user_input.lower() == 'quit':
           break
       
       messages.append({"role": "user", "content": user_input})
       print("\nGPT: ", end="", flush=True)
       
       response = client.chat.completions.create(
           model="gpt-3.5-turbo",
           messages=messages,
           stream=True
       )
       
       assistant_message = ""
       for chunk in response:
           if chunk.choices[0].delta.content:
               print(chunk.choices[0].delta.content, end="", flush=True)
               assistant_message += chunk.choices[0].delta.content
       
       if assistant_message:
           messages.append({"role": "assistant", "content": assistant_message})
       print()

2. Generate Data Examples:

   a. Quick Generation generate_data.py:

   b. Batch Generation generate_dataset.py:

   • Uses YAML configuration
   • Supports async processing
   • Handles large-scale generation

   Example configuration config/default_config.yaml:

   Run the generator:

   python examples/generate_dataset.py

   Key differences:

   • generate_data.py: Quick testing and small datasets
   • generate_dataset.py: Production use with:
     • Configuration management
     • Async processing
     • Batch generation
     • Flexible parameter control

   c. Classification Example classification.py:

   Features:

   • Automatic Prompt Optimization: Automatically optimize prompts through multiple training rounds
   • Accuracy Feedback: Calculate and display accuracy for each training round
   • Format Standardization: Automatically standardize model output formats
   • Progressive Learning: Support multi-round iterative training to improve effectiveness

   Run Examples:

   python examples/classification.py

   Output Example:

   开始第 1 轮训练...
   训练准确率: {'sentiment_accuracy': 0.4}
   新提示词效果更好: 1.0 > 0.4
   
   开始第 2 轮训练...
   训练准确率: {'sentiment_accuracy': 1.0}
   
   开始第 3 轮训练...
   训练准确率: {'sentiment_accuracy': 1.0}
   
   优化后的提示词:
   ```
   对商品评论进行情感分类。
   输入模板: 评论文本: {text}
   输出模板: 情感分类: {sentiment}
   可用标签: {'sentiment': ['正面', '负面', '中性']}
   ```
   

Common Issues

1. API Connection Errors:
   • Verify your API key is correct
   • Check if you need to use a proxy (add to .env):

     OPENAI_PROXY=http://127.0.0.1:7890
     

   • Ensure your API base URL is correct

Contributing 🤝

We welcome contributions from the community! If you'd like to contribute, please follow these steps:

1. Fork the repository.
2. Create a new branch (git checkout -b feature/YourFeature).
3. Commit your changes (git commit -m 'Add some feature').
4. Push to the branch (git push origin feature/YourFeature).
5. Open a pull request.

License 📄

This project is licensed under the MIT License - see the LICENSE file for details.

Contact 📬

For any questions or feedback, please reach out to us at bingzhenli@hotmail.com.

Project References 🔍

This project draws inspiration from the Adala framework, incorporating several key architectural concepts:

Core Components Inspired by Adala

1. Agent-Based Architecture

   • Autonomous data generation agents
   • Iterative learning capabilities
   • Environment-aware processing

2. Runtime System

   • Flexible model integration
   • Configurable execution environments
   • Support for multiple LLM providers

3. Environment Management

   • Ground truth dataset handling
   • Feedback collection mechanisms
   • Performance metrics tracking

4. Skills System

   • Task-specific capabilities
   • Customizable output formats
   • Validation mechanisms

Key Improvements

We've enhanced several aspects while adapting Adala's concepts:

• Simplified Configuration: Streamlined setup process with YAML-based configs
• Async Processing: Added robust async support for batch operations
• Enhanced Validation: Improved data quality checks and error handling
• Metrics Collection: Added comprehensive generation metrics and monitoring

Future Integrations

Planned features inspired by Adala:

• Memory management for long-term learning
• Advanced feedback collection mechanisms
• Multi-modal data generation support
• Enhanced runtime optimization

For more details about Adala's architecture, see their documentation.

🗺 Future Milestones

1. Enhanced Agent Architecture

• Implement teacher-student model architecture (using stronger models to guide weaker ones)
• Support multiple skill combinations and collaboration
• Add long-term memory management mechanism
• Implement asynchronous feedback collection

2. Environment System Enhancement

• Support human-in-the-loop feedback mechanisms
• Add real-time environment interaction capabilities
• Implement dynamic dataset management
• Support incremental learning scenarios

3. Runtime Optimization

• Support multiple LLM providers (e.g., Claude, Wenxin, etc.)
• Implement automatic model performance evaluation
• Add model call cost tracking
• Support batch processing optimization

4. Skill System Extension

• Add Named Entity Recognition (NER) skills
• Implement multi-task parallel learning
• Support cross-language skill transfer
• Add text generation skill templates

5. Toolchain Development

• Provide command-line tools
• Implement REST API interfaces
• Support Jupyter Notebook integration
• Add Web UI interfaces

6. Monitoring and Evaluation

• Implement detailed metrics collection system
• Add performance visualization panels
• Support experiment comparison analysis
• Implement automated testing frameworks

7. Multi-modal Support

• Add image processing capabilities
• Support speech input and output
• Implement cross-modal skills
• Support video content processing