DuckDB-NSQL-7B-v0.1: Specialized Text-to-SQL Generation

DuckDB-NSQL-7B-v0.1 is a 7 billion parameter model from the NSQL family, specifically engineered for advanced SQL generation tasks. Built upon Meta's Llama-2 7B, this model undergoes a unique training regimen that includes pre-training on a broad dataset of general SQL queries, followed by fine-tuning on 200,000 synthetically generated DuckDB text-to-SQL pairs. These DuckDB-specific pairs were created using Mixtral-8x7B-Instruct-v0.1 and guided by DuckDB v0.9.2 documentation, alongside transpiled SQL from the NSText2SQL dataset.

Key Capabilities

• DuckDB-Specific SQL Generation: Optimized to produce valid DuckDB SQL from natural language questions and database schemas.
• Comprehensive SQL Support: Unlike many text-to-SQL models, it can generate a full range of DuckDB SQL statements, not just SELECT queries, including those for official DuckDB extensions.
• Schema-Aware Querying: Effectively utilizes provided database schemas to generate accurate and contextually relevant SQL.

Good For

• Developers and data analysts needing to automate DuckDB SQL query generation from natural language.
• Applications requiring dynamic creation of complex DuckDB statements, including CREATE TABLE, INSERT, or other DDL/DML operations.
• Integrating natural language interfaces with DuckDB databases for enhanced accessibility and efficiency.