vishnurchityala/sql-gemma3
vishnurchityala/sql-gemma3 is a 1 billion parameter language model fine-tuned from Gemma 3 1B Instruct, specifically optimized for text-to-SQL generation. It was trained on a balanced subset of the Gretel synthetic_text_to_sql dataset to convert natural language questions and table schemas into SQL queries. This model excels at generating SQL from schema-aware prompts, making it suitable for learning, experimentation, and prototyping NL-to-SQL assistants.
Loading preview...
SQL-Gemma3: Text-to-SQL Generation
vishnurchityala/sql-gemma3 is a specialized 1 billion parameter model, fine-tuned from Gemma 3 1B Instruct, designed for efficient text-to-SQL generation. Its core capability lies in translating natural language questions and provided table schemas into SQL queries.
Key Capabilities
- SQL Generation: Converts schema-aware natural language prompts into SQL queries.
- Base Model: Built upon
unsloth/gemma-3-1b-it, leveraging its foundational language understanding. - Training Data: Utilizes a balanced sampled subset of the
gretelai/synthetic_text_to_sqldataset, focusing on improving SQL generation accuracy. - Performance Metrics: Achieved a reported training loss of
0.201and a test loss of0.21during its fine-tuning process.
Intended Use Cases
This model is particularly well-suited for:
- Prototyping: Rapid development of natural language to SQL assistants.
- Learning & Experimentation: Exploring text-to-SQL workflows and model behavior.
- Schema-Aware Querying: Generating SQL queries where table schemas are provided as context.
Limitations
It's important to note that the model's performance is currently summarized by loss metrics, not execution accuracy. The quality of generated SQL is highly dependent on the clarity of the provided schema and the format of the prompt. Users should review all generated queries, as the model may produce dialect-specific or invalid SQL in certain scenarios, and it is not guaranteed to generate correct, executable, or secure SQL for all prompts.