Script Planner Assistant▸ 
AI Apps 
AI News 
DuckDB-NSQL: Text-to-SQL Model Powered by LLMs
DuckDB-NSQL: Text-to-SQL Model Powered by LLMsIntroduction
DuckDB-NSQL, a text-to-SQL model specifically designed for the DuckDB database. You can describe your needs using natural language, and it will automatically convert them into SQL code—essentially allowing you to chat with your database using natural language.
DuckDB-NSQL greatly simplifies the database query process, making it easy for users who are not familiar with the SQL language to interact with databases and process data.
The DuckDB-NSQL-7B model is trained on approximately 200,000 synthetically generated and validated DuckDB SQL queries, as well as over 250,000 general text-to-SQL questions from Numbers Station.
DuckDB-NSQL can not only generate useful DuckDB code snippets but also generate SQL queries for answering analytical questions.
Hugging Face: https://huggingface.co/spaces/motherduckdb/DuckDB-NSQL-7B
Key Features of DuckDB-NSQL:
1. Natural Language Processing Capability: Able to understand and process natural language input, converting data query needs described in everyday language into SQL query code.
2. Optimized for DuckDB: Custom-built for the DuckDB-NSQL database, fully leveraging DuckDB's features and capabilities.
3. Efficient Query Generation: For common data query tasks such as creating tables, selecting data, sorting, and filtering, it can quickly generate accurate SQL code.
4. User-Friendly Interaction: Users do not need an in-depth understanding of SQL syntax; they can perform complex data queries simply by describing them in natural language.
5. Documentation-Style Query Guidance: The model's knowledge covers all features documented in DuckDB 0.9.2, including official extensions, functioning similar to a readily available documentation query tool.
6. Low Latency: To provide low-latency SQL assistance, the model uses a relatively small model size, making inference faster and more cost-effective.
7. Wide Range of Application Scenarios: Not only can it generate code snippets for DuckDB-NSQL, but it can also generate SQL queries for answering analytical questions.
8. Open Source and Easily Accessible: The model weights are fully public on Hugging Face, making it convenient for users to download and use.
9. Local Runtime Support: Supports running fully locally with llama.cpp, with complete instructions provided for local execution.
