rajaykumar12959/gemma-2-2b-sql-finetuned

Model Overview

This model, developed by rajaykumar12959, is a Gemma-2-2B architecture fine-tuned using QLoRA for Text-to-SQL generation. It converts natural language questions and database schemas into SQL queries, leveraging the gretelai/synthetic_text_to_sql dataset for training. The fine-tuning process was optimized for speed using Unsloth and Huggingface's TRL library.

Key Capabilities

• Generates complex SQL queries: Handles multi-table JOINs (INNER, LEFT, RIGHT), aggregation functions (SUM, COUNT, AVG, MIN, MAX), GROUP BY, HAVING clauses, and complex WHERE conditions.
• Supports advanced SQL operations: Capable of generating subqueries, Common Table Expressions (CTEs), date/time operations, and string functions.
• Efficient training: Fine-tuned with QLoRA, enabling efficient training on a single GPU with 8GB+ VRAM.

Good for

• Automating SQL query generation: Ideal for applications requiring the conversion of natural language into executable SQL.
• Database interaction tools: Can be integrated into tools that allow users to query databases using plain English.
• Educational and research purposes: Provides a robust foundation for exploring Text-to-SQL capabilities and further fine-tuning.

Limitations

While proficient, the model's current training (100 steps for demo) suggests that increasing max_steps to 300+ is recommended for production use. It primarily supports standard SQL syntax and has a maximum sequence length of 2048 tokens. Highly complex nested queries may require additional fine-tuning.