nl2sql-pretrained

🚀 MySQL Query Generator - From Scratch

A state-of-the-art GPT-style transformer model trained completely from scratch for natural language to MySQL query generation. This model demonstrates that high-quality language models can be built without relying on pre-trained weights, achieving excellent performance with a compact architecture.

🎯 Model Overview

This model specializes in converting natural language descriptions into syntactically correct MySQL queries. It was trained entirely from scratch using a custom transformer architecture, making it highly optimized for SQL generation tasks.

Key Features

• 🔧 Built from Scratch: No pre-trained weights - pure end-to-end training
• 💾 Lightweight: Compact 29.8M parameters for efficient deployment
• ⚡ High Performance: Excellent convergence with minimal overfitting
• 🎯 MySQL Optimized: Specifically tuned for MySQL syntax and patterns
• 📊 Production Ready: Robust performance across diverse query types

🏗️ Architecture

🎯 Performance Metrics

📊 Training Configuration

• Framework: PyTorch
• Optimizer: AdamW with weight decay
• Learning Rate Scheduler: CosineAnnealingLR
• Training Epochs: 8
• Training Examples: 24,293 high-quality samples
• Hardware: NVIDIA RTX 5080 16GB

📚 Dataset

The model was trained on a carefully curated dataset of 24,293 high-quality examples sourced from:

• 🔧 Synthetic SQL Queries: Custom-generated queries covering diverse MySQL patterns
• 🕷️ Spider Dataset: Complex multi-table queries with natural language descriptions
• 📖 WikiSQL Dataset: Real-world table-question pairs adapted for MySQL

All queries were specifically optimized for MySQL syntax and best practices, ensuring production-ready output.

🚀 Usage

This model excels at converting natural language descriptions into syntactically correct MySQL queries. Perfect for:

• Database query assistants
• Business intelligence tools
• Educational SQL learning platforms
• Automated report generation

Example Queries

# Basic Selection
"Show me all customers from New York"
# → SELECT * FROM customers WHERE city = 'New York';

# Aggregation
"Find total sales for each product"
# → SELECT product_name, SUM(sales) FROM sales_table GROUP BY product_name;

# Conditional Filtering
"List employees with salary greater than 50000"
# → SELECT * FROM employees WHERE salary > 50000;

📁 Model Files

🔬 Technical Details

Model Capabilities

• Multi-table Joins: Handles complex relationships between tables
• Aggregation Functions: SUM, COUNT, AVG, MIN, MAX operations
• Conditional Logic: WHERE clauses with AND/OR operators
• Sorting & Grouping: ORDER BY and GROUP BY operations
• Subqueries: Nested query generation for complex requirements

Limitations

• Optimized specifically for MySQL syntax (may not work with other SQL dialects)
• Best performance on queries similar to training data patterns
• May require fine-tuning for highly specialized domain vocabularies

📖 Citation

If you use this model in your research or applications, please cite:

@misc{mysql-query-generator-from-scratch,
  title={MySQL Query Generator: A GPT-style Transformer Trained From Scratch},
  author={Anonymous},
  year={2025},
  howpublished={\\url{https://huggingface.co/karthik-2905/nl2sql-pretrained}},
  note={Natural Language to SQL Query Generation}
}

📄 License

This model is released under the Apache 2.0 License, allowing for both commercial and non-commercial use.

🤝 Community & Support

• Open Source: Community-driven development
• Contributions Welcome: Feel free to submit improvements
• Issues & Feedback: Report problems or suggest enhancements
• Educational Use: Perfect for learning NL2SQL concepts

---

⭐ If you find this model useful, please give it a star and share it with others!