jasonvan

/

llama-2-13b-chat-text2sql-warmup-300

like 0

For example I am using following model to predict sql query with streamlit.

if st.button("Predict"):

    tok_model = "juierror/flan-t5-text2sql-with-schema"
    tokenizer,model = tokmod(tok_model)
    
    # text = "what interface is measure indicator code = 72_HR_ABX and version is 1 and source is TD"
    table_name = "df_qna"
    table_column = ['Patient_Name', 'Country', 'Disease', 'CUI', 'Snomed', 'Oxygen_Rate','Med_Type', 'Admission_Date']
    
    txt_sql = inference(question=sql_txt, table=table_column)
    
    
    ### SQL Modification
    sql_avg = ["AVG","COUNT","DISTINCT","MAX","MIN","SUM"]
    txt_sql = txt_sql.replace("table",table_name)
    sql_quotes = []
    for match in re.finditer("=",txt_sql):
        new_txt = txt_sql[match.span()[1]+1:]
        try:
            match2 = re.search("AND",new_txt)
            sql_quotes.append((new_txt[:match2.span()[0]]).strip())
        except:
            sql_quotes.append(new_txt.strip())
    
    for i in sql_quotes:
        qts = "'" + i + "'"
        txt_sql = txt_sql.replace(i, qts)
        
    for r in sql_avg:
        if r in txt_sql:
            rr = re.search(rf"{r} (\w+)", txt_sql)
            init = " " + rr[1]
            qts = "(" + rr[1] + ")"
            txt_sql = txt_sql.replace(init,qts)
        else:
            pass
        
        
    st.success(f"{txt_sql}")
    all_students = sqldf(txt_sql)
    
    st.text("Output - ")
    st.write(all_students)