T5 Text-to-SQL Conversion

This project utilizes the T5 base model for the task of converting natural language queries into SQL queries.

Training Process

• Fine-Tuning Parameters: All parameters of the T5 base model are fine-tuned during training.
• Epochs: Training is conducted over 20 epochs.
• Early Stopping: Early stopping is applied using validation and training loss to prevent overfitting.
• Evaluation Metrics: During training, the model is evaluated using ROUGE and BLEU metrics.
• Main Evaluation Metrics: For testing, ROUGE and BLEU metrics are primarily employed, along with query accuracy and query execution accuracy.

Model Performance

The chosen model achieved the following results:

• ROUGE Scores:

  • rouge1: 0.9601
  • rouge2: 0.9212
  • rougeL: 0.9459
  • rougeLsum: 0.9461

• BLEU Scores:

  • bleu: 0.8971
  • precisions:
    • 0.9549
    • 0.9217

• Test Loss: 0.0573

• Execution Accuracy: 0.7778

• Query Accuracy: 0.7094

These metrics highlight the effectiveness of the model in accurately converting natural language queries to SQL queries, showcasing its potential for practical applications in database interaction and information retrieval tasks.