|
|
--- |
|
|
library_name: transformers |
|
|
tags: [ |
|
|
text-to-sql |
|
|
] |
|
|
--- |
|
|
|
|
|
# Llama 2-7B Fine-Tuned for Text-to-SQL |
|
|
|
|
|
This model is a fine-tuned version of the **Llama 2-7B** model, specifically adapted for **Text-to-SQL** tasks. The model was trained to generate SQL queries from natural language questions, providing a robust solution for systems that need to translate user queries into executable SQL code. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Model Name**: Llama 2-7B Fine-Tuned for Text-to-SQL |
|
|
- **Base Model**: Llama 2-7B |
|
|
- **Model Developers**: Fine-tuned by MertML |
|
|
- **License**: Custom commercial license. Please refer to the repository for terms. |
|
|
- **Intended Use**: Designed for generating SQL queries from natural language input. Ideal for applications in databases, conversational agents, and data analysis tools. |
|
|
|
|
|
## Model Architecture |
|
|
|
|
|
Llama 2-7B is an autoregressive language model based on the transformer architecture. The fine-tuned version has been specifically adapted for the Text-to-SQL task, trained to convert user-written questions into valid and executable SQL queries using supervised fine-tuning. |
|
|
|
|
|
## Intended Use Cases |
|
|
Translating natural language queries into SQL queries, suitable for database query generation, business intelligence applications, and conversational agents that interact with databases. |
|
|
|
|
|
### Out-of-Scope Uses |
|
|
While this model is capable of text generation, it is fine-tuned specifically for Text-to-SQL tasks and may not perform well for general-purpose language generation tasks. |
|
|
|
|
|
## Training Data |
|
|
|
|
|
The model was fine-tuned using the [**refined-sql-create-context**](https://huggingface.co/datasets/MertML/refined-sql-create-context) dataset, which contains natural language queries, corresponding table schemas, and the correct SQL queries. This dataset was preprocessed to ensure that all queries were valid and executable on a MySQL database. |
|
|
|
|
|
- **Training Data Size**: 11,632 samples, split into training, validation, and test sets (80%, 10%, 10%). |
|
|
- **Data Source**: SQL-create-context dataset (refined for this task). |
|
|
- **Data Preprocessing**: Ambiguities in table schemas were resolved, invalid SQL queries were removed, and normalization was performed on SQL formatting for consistent evaluation. |
|
|
|
|
|
## Model Performance |
|
|
|
|
|
The fine-tuned Llama 2-7B on Text-to-SQL demonstrated significant improvements over the base model in generating syntactically correct and contextually relevant SQL queries. Performance was evaluated on a set of queries with varying levels of difficulty, and the model was benchmarked against the [**refined-sql-create-context**](https://huggingface.co/datasets/MertML/refined-sql-create-context) datasets. |
|
|
|
|
|
### Evaluation Metrics |
|
|
|
|
|
- **Accuracy**: Measures the percentage of generated SQL queries that are syntactically and semantically correct. |
|
|
- **Execution Success Rate**: Measures the percentage of SQL queries that execute successfully against a database. |
|
|
- **Response Quality**: Assesses the relevance and correctness of the generated SQL queries in context. |
|
|
|
|
|
|
