AuricErgeson/Antelope-textTosql
AuricErgeson/Antelope-textTosql is a 3 billion parameter causal language model developed by Auric Ergeson Nitonde, based on Microsoft's Phi-2 architecture. This lightweight model is specifically fine-tuned for converting natural language questions into SQL queries, making it suitable for CPU inference. It excels at generating SQL from plain English across various database schemas, trained on the Spider dataset with over 7,000 examples.
Loading preview...
Antelope Text-to-SQL Overview
AuricErgeson/Antelope-textTosql is a specialized 3 billion parameter model designed to convert natural language questions into SQL queries. Built upon the microsoft/phi-2 base model and fine-tuned using QLoRA, it offers a lightweight solution for generating SQL from plain English. The model is trained on the extensive Spider dataset, encompassing over 7,000 examples across more than 200 diverse database schemas, enabling cross-domain applicability.
Key Capabilities & Features
- Text-to-SQL Conversion: Translates user questions like "How many employees are there?" into accurate SQL queries such as
SELECT COUNT(*) FROM employees. - Lightweight & Fast: With 2.7 billion parameters, it is designed to run efficiently on modest hardware, including CPU inference, making it accessible for various deployment scenarios.
- Easy Integration: Works out-of-the-box with the
transformerslibrary, simplifying implementation. - Open License: Released under the MIT license, allowing for broad usage.
When to Use This Model
This model is ideal for applications requiring efficient and accurate conversion of natural language to SQL, particularly when:
- You need to enable users to query databases without SQL expertise.
- Resource constraints necessitate a smaller, faster model capable of CPU inference.
- Your use case involves a variety of database schemas, benefiting from its cross-domain training.
Limitations
It's important to note that the model is not recommended for production databases without output validation and may struggle with highly complex multi-join queries. It also requires the database name to be provided for optimal results, as it does not infer table/column names.