Quick start

This repository uses git submodules. Clone it like this:

$ git clone git@github.com:JiexingQi/picard.git
$ cd picard
$ git submodule update --init --recursive

Requirements

Suggested environment to run the code:

python 3.9.7

You can make a new conda envirment using:

conda create -n picard python==3.9.7

And then, you may need to install these packages using pip：

• sqlparse==0.4.2
• nltk==3.6.5
• wandb==0.12.7
• transformers==4.13.0
• datasets==1.16.1
• tenacity==8.0.1
• rapidfuzz==1.8.3

or using requirements.txt

Run code

First, difine a config file in /configs, and then use the command to run the code(in this example, the config file is train_0125_example.json):

CUDA_VISIBLE_DEVICES="2,3" python3 -m torch.distributed.launch --nnodes=1 --nproc_per_node=2 seq2seq/run_seq2seq.py configs/train_0125_example.json

Note

• You should set --nproc_per_node=#gpus to --nproc_per_node=2 make full use of all gpus.
• A recommand total_batch_size = #gpus * gradient_accumulation_steps * per_device_train_batch_size is 2048.

Config file

In config json file, you must set the correct filepath for relation filepath.

"lge_relation_path" : "/home/jxqi/text2sql/data"

this key-value pair set the relation filepath.

Relation file

The relation files are aviliable in Google drive: https://drive.google.com/drive/folders/1cads4MN02FUj5gUwcwP6mYzrSNkNWD9l?usp=sharing