| ---
|
| license: cc-by-sa-4.0
|
| language:
|
| - en
|
| base_model:
|
| - defog/sqlcoder-7b-2
|
| tags:
|
| - text-generation
|
| - text-to-sql
|
| - sql
|
| - coding
|
| - database
|
| - instruction-following
|
| - efficient-model
|
| ---
|
|
|
| ## SQLCoder-7B-2
|
|
|
| SQLCoder-7B-2 is a specialized code generation model developed by Defog and optimized for Text-to-SQL generation, database querying, and structured data analysis workflows. This repository contains GGUF quantized variants of the model optimized for efficient local inference using llama.cpp.
|
|
|
| The model is designed to translate natural language questions into syntactically correct SQL queries across a variety of database schemas. The quantized formats significantly reduce memory requirements while preserving strong SQL generation capability, enabling practical deployment on consumer hardware and edge environments.
|
|
|
| ---
|
|
|
| ## Model Overview
|
|
|
| - **Model Name**: SQLCoder-7B-2
|
| - **Base Model**: defog/sqlcoder-7b-2
|
| - **Architecture**: Decoder-only Transformer
|
| - **Parameter Count**: 7 Billion
|
| - **Modalities**: Text
|
| - **Primary Languages**: English
|
| - **Developer**: Defog
|
| - **License**: CC BY-SA 4.0
|
|
|
| ---
|
|
|
| ## Quantization Formats
|
|
|
| This repository provides various GGUF quantized versions of the SQLCoder-7B-2 model, optimized for efficient local inference using `llama.cpp`. Below are the details of the available I-Matrix (IQ) formats.
|
|
|
| ### IQ3_M
|
|
|
| - Size reduction of approx 76.89% (2.90 GB) compared to 16-bit (12.55 GB)
|
| - Aggressive 3-bit quantization optimized for maximum memory reduction
|
| - Suitable for low-memory systems and CPU-based inference
|
| - Enables lightweight deployment of Text-to-SQL workflows on constrained hardware
|
| - SQL generation quality may degrade on highly complex schemas, multi-table joins, and advanced analytical queries
|
|
|
| ### IQ4_NL
|
|
|
| - Size reduction of approx 71.47% (3.58 GB) compared to 16-bit (12.55 GB)
|
| - Advanced 4-bit non-linear quantization designed to better preserve SQL generation quality
|
| - More suitable for structured query generation, schema understanding, and analytical database workflows
|
| - Typically provides stronger consistency and reduced quantization loss compared to lower-bit formats
|
| - Slightly increased computational overhead during inference
|
|
|
| ### IQ4_XS
|
|
|
| - Size reduction of approx 72.91% (3.40 GB) compared to 16-bit (12.55 GB)
|
| - Balanced 4-bit quantization focused on efficiency and stable SQL generation performance
|
| - Good trade-off between model size, inference speed, and query quality
|
| - Suitable for general-purpose Text-to-SQL applications and local database assistants
|
| - Maintains reliable query generation behavior across most practical workloads
|
|
|
| ---
|
|
|
| ## Training Background (Original Model)
|
|
|
| SQLCoder-7B-2 is trained with a focus on natural language to SQL translation, database reasoning, schema understanding, and structured query generation.
|
|
|
| ### Pretraining
|
|
|
| - Large-scale language model pretraining across diverse textual and code datasets
|
| - Focus on programming, structured reasoning, and database-related knowledge
|
| - Optimized for downstream SQL generation and analytical workflows
|
|
|
| ### Instruction Tuning
|
|
|
| - Refined using Text-to-SQL datasets and schema-aware instruction tuning
|
| - Enhanced for generating syntactically correct SQL queries
|
| - Improved consistency for database querying, filtering, aggregation, and analytical operations
|
|
|
| ---
|
|
|
| ## Key Capabilities
|
|
|
| - **Text-to-SQL Generation**
|
| Converts natural language requests into executable SQL queries.
|
|
|
| - **Database Schema Understanding**
|
| Interprets table structures, relationships, and schema context effectively.
|
|
|
| - **SQL Query Construction**
|
| Supports filtering, joins, aggregations, grouping, ordering, and analytical operations.
|
|
|
| - **Efficient Local Deployment**
|
| Quantized variants enable practical offline inference on consumer hardware.
|
|
|
| - **Structured Data Analysis**
|
| Assists with database exploration and business intelligence workflows.
|
|
|
| ---
|
|
|
| ## Usage Example
|
|
|
| ### Using llama.cpp
|
|
|
| ```bash
|
| ./llama-cli \
|
| -m SandlogicTechnologies/sqlcoder-7b-2_IQ4_NL.gguf \
|
| -p "Generate a SQL query to find the top 10 customers by total revenue."
|
| ```
|
|
|
| ---
|
|
|
| ## Recommended Usecases
|
|
|
| - **Text-to-SQL Applications**
|
| Build natural language interfaces for relational databases.
|
|
|
| - **Database Assistants**
|
| Generate and explain SQL queries from user requests.
|
|
|
| - **Business Intelligence Workflows**
|
| Assist with reporting, aggregation, and analytical query generation.
|
|
|
| - **Data Exploration**
|
| Simplify interaction with structured databases through natural language.
|
|
|
| - **Research and Experimentation**
|
| Evaluate Text-to-SQL performance and database reasoning workflows.
|
|
|
| ---
|
|
|
| ## Acknowledgments
|
|
|
| These quantized models are based on the original work by the **Defog** development team.
|
|
|
| Special thanks to:
|
|
|
| - The Defog team for developing and releasing the SQLCoder-7B-2 model.
|
|
|
| - **Georgi Gerganov** and the `llama.cpp` open-source community for enabling efficient quantization and inference via the GGUF format.
|
|
|
| ---
|
|
|
| ## Contact
|
|
|
| For questions, feedback, or support, please reach out at [support@sandlogic.com](mailto:support@sandlogic.com) or visit [https://www.sandlogic.com/](https://www.sandlogic.com/)
|
| |
