Mistral-7B-Text2SQL

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.3 on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 0.4643

Model description

This repository contains a fine-tuned version of the Mistral 7B model, tailored specifically for text-to-SQL tasks. The model is designed to convert natural language queries into structured SQL queries, enabling seamless interaction with databases through conversational language.

Intended uses & limitations

The Mistral-7B-Text2SQL model is intended for applications that require converting natural language queries into SQL commands. Suitable use cases include:

Conversational Agents: Allowing users to retrieve information from databases through natural language interaction. Data Analytics: Enabling non-technical users to query databases without needing to know SQL syntax. Business Intelligence: Supporting decision-making processes by simplifying data access.

Training and evaluation data

The model was fine-tuned using the generator dataset, which consists of a variety of natural language queries paired with corresponding SQL commands. The dataset is designed to cover a wide range of query types, allowing the model to generalize better across different types of SQL queries.

Dataset Characteristics Diversity: The dataset includes examples from various domains, ensuring that the model learns to handle a broad spectrum of queries. Size: (Include the size of the dataset, e.g., the number of examples if available.) Annotations: Each example includes natural language input along with the expected SQL output, facilitating supervised learning.

Training results

Training Loss	Epoch	Step	Validation Loss
1.8346	0.4	10	0.7031
0.5882	0.8	20	0.5273
0.487	1.2	30	0.4850
0.4423	1.6	40	0.4675
0.4235	2.0	50	0.4564
0.3464	2.4	60	0.4690
0.3411	2.8	70	0.4643

Framework versions

• PEFT 0.13.2
• Transformers 4.45.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1