Chat2DB

Pay Attention

Live demo

You can access the live demo version of Chat2DB by visiting 8.138.157.109. Please be aware that due to the high cost of GPU, the live website currently only supports engines based on deepseek for LLM and t5-base for PLM. If you wish to experience the full version, we have provided a comprehensive multi-engine Chat2DB framework code in our repository. If you wish to experience the full version's features, please prepare an A800 GPU locally and deploy it according to the following instructions.

Video demo

You can watch the live video of Chat2DB by visiting Chat2DB-demonstration

Preview

Overview

Description

Code implementation for the paper Chat2DB: Chatting to the Database with Interactive Agent Assisted Language Models.

Prerequisites

Create a virtual anaconda environment and install the required modules and tools:

conda create -n chat2db python=3.10
conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3 -c pytorch
pip install -r requirements.txt
pip install https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-2.2.0/en_core_web_sm-2.2.0.tar.gz
python nltk_downloader.py

Create several folders:

mkdir checkpoints
mkdir data
mkdir logs

Prepare Data

Download the data here and unzip them in the data/ folder as follows:

data/
├── spider/
│   ├── database/
│   ├── ...
|   └── tables.json
└── usr/
    └── database/

Prepare Model

Download the Embeddings Models and put it into the checkpoint folders.
Download the Schema Filter Models and put it into the checkpoint folders.
For the parsers, due to our company's privacy protection policies, we have made some publicly available parsers available for download below. We also welcome you to use your own trained parsers.

Parsers Checkpoint Links

t5-3b text2sql-t5-3b

t5-large text2sql-t5-large

t5-base text2sql-t5-base

deepseek-6.7b deepsql-6.7b

Parsers	Checkpoint Links
t5-3b	text2sql-t5-3b
t5-large	text2sql-t5-large
t5-base	text2sql-t5-base
deepseek-6.7b	deepsql-6.7b

After download these checkpoints, you need to change the settings.json and modelend/utils/args.py to adapt these parsers. Please pay special attention to the folder names of the checkpoint which must match those in the configuration file.

Deploy & Serve

Grant the execution permission to the script.

chmod +x ./run_all.sh

Starting the Chat2DB Service.

./run_all.sh start

At this moment, the backend server, model side, and frontend visualization interface of Chat2DB will each start in the background and listen on ports 8299, 8091, and 8080 respectively. Of course, you can modify these in settings.py and run_all.sh.

You will find three service output logs in the logs folder, which is organized as follows. You can monitor your services by observing the log outputs.

logs/
├── backend.log
├── frontend.log
└── modelend.log

Stopping all the Chat2DB Service:

./run_all.sh stop

Acknowledgements

We would thanks to RESDSQL(paper, code), DTS-SQL(paper, code), Spider(paper, dataset) for their interesting work and open-sourced code and dataset.

kanseaveg/Chat2DB