NLSQL / app.py
HusnaManakkot's picture
Update app.py
04901e7 verified
import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from datasets import load_dataset
# Load the WikiSQL dataset
wikisql_dataset = load_dataset("wikisql", split='train') # Load a subset of the dataset
# Extract schema information from the WikiSQL dataset
table_names = set()
column_names = set()
for item in wikisql_dataset:
table_names.add(item['table']['name'])
for column in item['table']['header']:
column_names.add(column)
# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-wikiSQL")
model = AutoModelForSeq2SeqLM.from_pretrained("mrm8488/t5-base-finetuned-wikiSQL")
def generate_sql_from_user_input(query):
# Generate SQL for the user's query
input_text = "translate English to SQL: " + query
inputs = tokenizer(input_text, return_tensors="pt", padding=True)
outputs = model.generate(**inputs, max_length=512)
sql_query = tokenizer.decode(outputs[0], skip_special_tokens=True)
# Post-process the SQL query to match the dataset's schema
for table_name in table_names:
if "TABLE" in sql_query:
sql_query = sql_query.replace("TABLE", table_name)
break # Assuming only one table is referenced in the query
for column_name in column_names:
if "COLUMN" in sql_query:
sql_query = sql_query.replace("COLUMN", column_name, 1)
return sql_query
# Create a Gradio interface
interface = gr.Interface(
fn=generate_sql_from_user_input,
inputs=gr.Textbox(label="Enter your natural language query"),
outputs=gr.Textbox(label="Generated SQL Query"),
title="NL to SQL with T5 using WikiSQL Dataset",
description="This model generates an SQL query for your natural language input based on the WikiSQL dataset."
)
# Launch the app
if __name__ == "__main__":
interface.launch()