from types import ClassMethodDescriptorType from click import prompt from dotenv import load_dotenv from langchain.llms import OpenAI from langchain import PromptTemplate from langchain.chains import LLMChain import streamlit as st from streamlit_extras.let_it_rain import rain import os import time import random #setting up the prerequites # load_dotenv() # os.environ["OPENAI_API_KEY"] = os.getenv('openai_key') llm = OpenAI(temperature = 0.8) LANGCHAIN_TRACING_V2="true" # os.environ["LANGCHAIN_API_KEY"] = os.getenv('LANGCHAIN_API_KEY') # os.environ["LANGCHAIN_TRACING_V2"] = "true" #llm2 = OpenAI(model_name = "gpt-4-turbo-preview", temperature = 0.7) st.set_page_config(page_title="Schem-SQL-Query", page_icon="🧊", layout="wide", initial_sidebar_state="collapsed") rain( emoji="❄️", font_size=20, falling_speed=10, animation_length="infinite", ) with open( "style.css" ) as css: st.markdown( f'' , unsafe_allow_html= True) st.markdown("

SQL Generator

", unsafe_allow_html=True) col1 , col2 = st.columns([0.4,0.6], gap="large") col1.header("Database Schema Input") schema_input = col1.text_area("Submit Database Schema", height=450) schema_submit_button = col1.button("Submit Schema", key="schema_submit") col2.header("Enter your question") user_prompt = col2.text_area("User's Prompt", height=150) prompt_submit_button = col2.button("Submit Prompt", key="prompt_submit") col2.markdown("---") #query prompt q_template = """You are an expert in writing SQL queries thus, based on the table schema below, write a SQL query that would answer the user's question: {schema} \n Question: {user_prompt} \nSQL Query:""" ## Prompt Templates first_input_prompt = PromptTemplate( input_variables = ['schema','user_prompt'], template = q_template ) #query chain chain1 = LLMChain(llm = llm, prompt = first_input_prompt, verbose=True, output_key='query') ######################## #scenario prompt s_template = """Based on the following schema, generate 9 sql questions that are relevant to the schema. The questions should be in increasing order of difficulty and complexity. Include atleast one problem which involves "join" operation. Just describe the problem statement in one line. Schema: {schema} """ third_input_prompt = PromptTemplate( input_variables = ['schema'], template = s_template ) #scenario chain chain3 = LLMChain(llm = llm, prompt = third_input_prompt, verbose=True, output_key='scenario') if schema_submit_button: with col1: with st.spinner('Wait for it...'): time.sleep(random.random() * 3) col1.success("Schema Submitted Successfully!!") expander = col1.expander("See Scenarios") scenarios = chain3({"schema": schema_input}, return_only_outputs=True) expander.write(scenarios["scenario"]) ############################ #algorithm prompt a_template = """You are an expert in explaining SQL queries thus Explain the following SQL Query in an Algorithmic step-by-step fashion. Write "Start" and "End" to denote the begining and ending of the Algorithm. Write at least 4 lines and a maximum of 12 lines to explain. Explain each step briefly, in a single line. Refer the below example. Example : Start 1. Start by selecting distinct customer details and product application details. 2. Join the customers table with the life events table on customer id. 3. Join the resulting set with the products table on customer_id. 4. Filter for customers who are household heads. 5. Filter for customers located in California. 6. Filter for life events that include 'relocation'. 7. Filter for life events that occurred in the last 6 months. 8. Filter for product applications that include 'insurance'. 9. Filter for product applications that occurred in the last 6 months. \n End Query: {query}""" second_input_prompt = PromptTemplate( input_variables = ['query'], template = a_template ) #algorithm chain chain2 = LLMChain(llm = llm, prompt = second_input_prompt, verbose=True, output_key='algorithm') if prompt_submit_button: with st.spinner('Generating response...'): qu = chain1({"schema": schema_input, "user_prompt":user_prompt}, return_only_outputs=True) #print(type(qu)) col2.subheader("Query") col2.code(qu["query"], language="sql") #st.write(qu["query"]) col2.markdown("---") col2.subheader("Algorithm") algo = chain2({"query": qu}, return_only_outputs=True) col2.write(algo["algorithm"])