sql_t5.py

import streamlit as st
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch
import os

def make_query(context, question):
    result_query = f'''You are a SQL expert with extensive experience, you need to create a query to answer the question.
    
### Database schema (PostgreSQL):
{context}

### Question:
{question}

### SQL Query: '''
    return result_query

st.set_page_config(page_title="SQL-to-Text with T5", page_icon="🤖")
st.title("SQL Query Generator with T5")

examples = [
    {
        "question": "Who was the music director in 1971 for the movie Kalyani?",
        "description": "CREATE TABLE table_name_7 (Music VARCHAR, year VARCHAR, movie__in_kannada__ VARCHAR)"
    },
    {
        "question": "What's the highest with a capacity of greater than 4,000 and an average of 615?",
        "description": "CREATE TABLE table_name_20 (highest INTEGER, average VARCHAR, capacity VARCHAR)"
    },
    {
        "question": "If the letters is φαν, what is the founding date?",
        "description": "CREATE TABLE table_2538117_7 (founding_date VARCHAR, letters VARCHAR)"
    },
    {
        "question": "How many weeks had a game on November 26, 1978, and an attendance higher than 26,248?",
        "description": "CREATE TABLE table_name_88 (week VARCHAR, date VARCHAR, attendance VARCHAR)"
    },
    {
        "question": "How many television service are in italian and n°is greater than 856.0?",
        "description": "CREATE TABLE table_15887683_15 (television_service VARCHAR, language VARCHAR, n° VARCHAR)"
    },
    {
        "question": "What date was Bury the home team?",
        "description": "CREATE TABLE table_name_67 (date VARCHAR, away_team VARCHAR)"
    },
    {
        "question": "What regular season result had an average attendance less than 942?",
        "description": "CREATE TABLE table_name_16 (reg_season VARCHAR, avg_attendance INTEGER)"
    },
    {
        "question": "What is the value for 2011 when `a` is the value for 2009, and 4r is the value for 2013?",
        "description": "CREATE TABLE table_name_89 (Id VARCHAR)"
    },
    {
        "question": "Who wrote episode with production code 1.01?",
        "description": "CREATE TABLE table_28089666_1 (written_by VARCHAR, production_code VARCHAR)"
    },
    {
        "question": "find the names of museums which have more staff than the minimum staff number of all museums opened after 2010.",
        "description": "CREATE TABLE museum (name VARCHAR, num_of_staff INTEGER, open_year INTEGER)"
    }
]

@st.cache_resource
def load_model():
    script_dir = os.path.dirname(os.path.abspath(__file__))
    model_path = os.path.join(script_dir, "model")
    
    if not os.path.exists(model_path):
        raise FileNotFoundError(f"Model directory not found at {model_path}")
    
    try:
        tokenizer = AutoTokenizer.from_pretrained("google-t5/t5-small")
        model = AutoModelForSeq2SeqLM.from_pretrained(model_path)
        model.eval()
        return model, tokenizer
    except Exception as e:
        raise RuntimeError(f"Error loading model: {str(e)}")

try:
    model, tokenizer = load_model()
except Exception as e:
    st.error(f"Failed to load model: {str(e)}")
    st.stop()

if 'current_description' not in st.session_state:
    st.session_state.current_description = """CREATE TABLE table_name_28 (played INTEGER, points VARCHAR, position VARCHAR)"""
if 'current_question' not in st.session_state:
    st.session_state.current_question = "Which Played has a Points of 2, and a Position smaller than 8?"

def load_example(example):
    st.session_state.current_description = example["description"]
    st.session_state.current_question = example["question"]

st.subheader("Примеры:")
cols = st.columns(2)
for i, example in enumerate(examples):
    col = cols[i % 2]
    if col.button(
        f"Пример {i+1}: {example['question'][:30]}...",
        key=f"example_{i}",
    ):
        load_example(example)
        st.rerun()

with st.form("query_form"):
    description = st.text_area(
        "Описание таблицы (столбцы и их типы):",
        st.session_state.current_description,
        height=150,
        key="desc_input"
    )
    
    question = st.text_input(
        "Ваш вопрос:", 
        st.session_state.current_question,
        key="question_input"
    )
    
    submitted = st.form_submit_button("Сгенерировать запрос")

if submitted:
    if description and question:
        input_text = make_query(description, question)
        try:
            input_ids = tokenizer.encode(input_text, return_tensors="pt")
            with st.spinner("Генерация запроса..."):
                outputs = model.generate(
                    input_ids,
                    max_length=200,
                    num_beams=5,
                    early_stopping=True,
                    pad_token_id=tokenizer.eos_token_id,
                )
                generated_sql = tokenizer.decode(outputs[0], skip_special_tokens=True)
                st.subheader("Результат:")
                st.code(generated_sql, language="sql")
        except Exception as e:
            st.error(f"Ошибка при генерации: {str(e)}")
    else:
        st.warning("Пожалуйста, заполните описание таблицы и вопрос")