evaluation/table_related_benchmarks/text2sql/src/gpt_request.py

#!/usr/bin/env python3
import argparse
import fnmatch
import json
import os
import pdb
import pickle
import re
import sqlite3
from typing import Dict, List, Tuple

import openai
import pandas as pd
import sqlparse
from tqdm import tqdm
from vllm import LLM, SamplingParams
from transformers import AutoTokenizer
import os
from openai import AzureOpenAI

def new_directory(path):  
    if not os.path.exists(path):  
        os.makedirs(path)  

def load_json(data_path):
    with open(data_path, "r") as f:
        datas = json.load(f)
    return datas

def get_db_schemas(bench_root: str, db_name: str) -> Dict[str, str]:
    """
    Read an sqlite file, and return the CREATE commands for each of the tables in the database.
    """
    asdf = 'database' if bench_root == 'spider' else 'databases'
    with sqlite3.connect(f'file:{bench_root}/{asdf}/{db_name}/{db_name}.sqlite?mode=ro', uri=True) as conn:
        # conn.text_factory = bytes
        cursor = conn.cursor()
        cursor.execute("SELECT name FROM sqlite_master WHERE type='table';")
        tables = cursor.fetchall()
        schemas = {}
        for table in tables:
            cursor.execute("SELECT sql FROM sqlite_master WHERE type='table' AND name='{}';".format(table[0]))
            schemas[table[0]] = cursor.fetchone()[0]

        return schemas

def nice_look_table(column_names: list, values: list):
    rows = []
    # Determine the maximum width of each column
    widths = [max(len(str(value[i])) for value in values + [column_names]) for i in range(len(column_names))]

    # Print the column names
    header = ''.join(f'{column.rjust(width)} ' for column, width in zip(column_names, widths))
    # print(header)
    # Print the values
    for value in values:
        row = ''.join(f'{str(v).rjust(width)} ' for v, width in zip(value, widths))
        rows.append(row)
    rows = "\n".join(rows)
    final_output = header + '\n' + rows
    return final_output

def generate_schema_prompt(db_path, num_rows=None):
    # extract create ddls
    '''
    :param root_place:
    :param db_name:
    :return:
    '''
    full_schema_prompt_list = []
    conn = sqlite3.connect(db_path)
    # Create a cursor object
    cursor = conn.cursor()
    cursor.execute("SELECT name FROM sqlite_master WHERE type='table'")
    tables = cursor.fetchall()
    schemas = {}
    for table in tables:
        if table == 'sqlite_sequence':
            continue
        cursor.execute("SELECT sql FROM sqlite_master WHERE type='table' AND name='{}';".format(table[0]))
        create_prompt = cursor.fetchone()[0]
        schemas[table[0]] = create_prompt
        if num_rows:
            cur_table = table[0]
            if cur_table in ['order', 'by', 'group']:
                cur_table = "`{}`".format(cur_table)

            cursor.execute("SELECT * FROM {} LIMIT {}".format(cur_table, num_rows))
            column_names = [description[0] for description in cursor.description]
            values = cursor.fetchall()
            rows_prompt = nice_look_table(column_names=column_names, values=values)
            verbose_prompt = "/* \n {} example rows: \n SELECT * FROM {} LIMIT {}; \n {} \n */".format(num_rows, cur_table, num_rows, rows_prompt)
            schemas[table[0]] = "{} \n {}".format(create_prompt, verbose_prompt)

    for k, v in schemas.items():
        full_schema_prompt_list.append(v)

    schema_prompt = "\n\n".join(full_schema_prompt_list)

    return schema_prompt

def generate_comment_prompt(question, knowledge=None):
    pattern_prompt_no_kg = "-- Using valid SQLite, answer the following questions for the tables provided above."
    pattern_prompt_kg = "-- Using valid SQLite and understading External Knowledge, answer the following questions for the tables provided above."
    # question_prompt = "-- {}".format(question) + '\n SELECT '
    question_prompt = "-- {}".format(question)
    knowledge_prompt = "-- External Knowledge: {}".format(knowledge)

    if not knowledge:
        result_prompt = pattern_prompt_no_kg + '\n' + question_prompt
    else:
        result_prompt = knowledge_prompt + '\n' + pattern_prompt_kg + '\n' + question_prompt

    return result_prompt

def cot_wizard():
    cot = "\nGenerate the SQL after thinking step by step: "
    # cot = "\nCarefully reason through each step to generate the SQL query:"
    return cot

def few_shot():
    ini_table = "CREATE TABLE singer\n(\n    singer_id         TEXT not null\n        primary key,\n    nation       TEXT  not null,\n    sname       TEXT null,\n    dname       TEXT null,\n    cname       TEXT null,\n    age    INTEGER         not null,\n    year  INTEGER          not null,\n    birth_year  INTEGER          null,\n    salary  REAL          null,\n    city TEXT          null,\n    phone_number   INTEGER          null,\n--     tax   REAL      null,\n)"
    ini_prompt = "-- External Knowledge: age = year - birth_year;\n-- Using valid SQLite and understading External Knowledge, answer the following questions for the tables provided above.\n-- How many singers in USA who is older than 27?\nThe final SQL is: Let's think step by step."
    ini_cot_result = "1. referring to external knowledge, we need to filter singers 'by year' - 'birth_year' > 27; 2. we should find out the singers of step 1 in which nation = 'US', 3. use COUNT() to count how many singers. Finally the SQL is: SELECT COUNT(*) FROM singer WHERE year - birth_year > 27;</s>"
    
    one_shot_demo = ini_table + '\n' + ini_prompt + '\n' + ini_cot_result
    
    return one_shot_demo

def few_shot_no_kg():
    ini_table = "CREATE TABLE singer\n(\n    singer_id         TEXT not null\n        primary key,\n    nation       TEXT  not null,\n    sname       TEXT null,\n    dname       TEXT null,\n    cname       TEXT null,\n    age    INTEGER         not null,\n    year  INTEGER          not null,\n    age  INTEGER          null,\n    salary  REAL          null,\n    city TEXT          null,\n    phone_number   INTEGER          null,\n--     tax   REAL      null,\n)"
    ini_prompt = "-- External Knowledge:\n-- Using valid SQLite and understading External Knowledge, answer the following questions for the tables provided above.\n-- How many singers in USA who is older than 27?\nThe final SQL is: Let's think step by step."
    ini_cot_result = "1. 'older than 27' refers to age > 27 in SQL; 2. we should find out the singers of step 1 in which nation = 'US', 3. use COUNT() to count how many singers. Finally the SQL is: SELECT COUNT(*) FROM singer WHERE age > 27;</s>"
    
    one_shot_demo = ini_table + '\n' + ini_prompt + '\n' + ini_cot_result
    
    return one_shot_demo



def generate_combined_prompts_one(db_path, question, knowledge=None):
    schema_prompt = generate_schema_prompt(db_path, num_rows=None) # This is the entry to collect values
    comment_prompt = generate_comment_prompt(question, knowledge)

    combined_prompts = schema_prompt + '\n\n' + comment_prompt + cot_wizard() + '\nSELECT '
    # combined_prompts = few_shot_no_kg() + '\n\n' + schema_prompt + '\n\n' + comment_prompt
    # print("="*100)
    # print(combined_prompts)
    # print("="*100)
    return combined_prompts


def quota_giveup(e):
    return isinstance(e, openai.error.RateLimitError) and "quota" in str(e)


def connect_gpt(engine, prompt, max_tokens, temperature, stop):
    try:
        result = openai.Completion.create(engine=engine, prompt=prompt, max_tokens=max_tokens, temperature=temperature, stop=stop)
    except Exception as e:
        result = 'error:{}'.format(e)
    return result

def llm_generate_result(model_name_or_path, gpus_num, prompt_ls, args=None):

    print("model", model_name_or_path)
    tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
    print(
        "load tokenizer {} from {} over.".format(
            tokenizer.__class__, model_name_or_path
        )
    )
    llm_args = {
        "model": model_name_or_path,
        "gpu_memory_utilization": 0.95,
        "trust_remote_code": True,
        "tensor_parallel_size": gpus_num,
        "dtype": "half",
        "max_model_len": 8192,
        "enforce_eager": True,
    }

    llm = LLM(**llm_args)
    sampling_params = SamplingParams(
        temperature=0,
        max_tokens=1024,
        top_p=0.95,
        stop_token_ids=[tokenizer.eos_token_id],
    )

    messages_list = []
    num = 0
    for prompt in tqdm(prompt_ls, desc="trans prompt"):
        message = [{"role": "user", "content": prompt}]
        messages_list.append(
            tokenizer.apply_chat_template(
                message, tokenize=False, add_generation_prompt=True
            )
        )
        tk = tokenizer.apply_chat_template(
                message, tokenize=True, add_generation_prompt=True
            )
        if len(tk) > 7168:
            print("="*100)
            # print(tk)
            num += 1
    # print("="*100, "cut nums: ", num)

    outputs = llm.generate(messages_list, sampling_params=sampling_params)
    generated_res = []
    ori_generated_res = []
    for i, output in enumerate(tqdm(outputs)):
        text = output.outputs[0].text
        ori_generated_res.append(text)
        sql = parser_sql(text)
        generated_res.append(sql)

    return generated_res, ori_generated_res


def gpt_generate_result(model_name_or_path, gpus_num, prompt_ls, args=None):

    client = AzureOpenAI(
    api_key=os.getenv("AZURE_OPENAI_API_KEY"),  
    api_version="2024-07-01-preview",
    azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"),
    max_retries=3
    )

    generated_res = []
    ori_generated_res = []

    output_name =  os.path.join(args.data_output_path, f'{args.eval_data_name}_{args.mode}_{args.is_use_knowledge}_temp.json')
    unparser_name = os.path.join(args.data_output_path,f'{args.eval_data_name}_{args.mode}_{args.is_use_knowledge}_unparser_temp.json')


    if os.path.exists(unparser_name):
        ori_generated_res_dict = load_json(unparser_name)
        generated_res_dict = load_json(output_name)

        generated_res = [v for k,v in generated_res_dict.items()]
        ori_generated_res = [v for k,v in ori_generated_res_dict.items()]

        
    for i in tqdm(range(len(prompt_ls))):
        if i < len(generated_res):
            continue
        prompt = prompt_ls[i]
        response = client.chat.completions.create(
                model="gpt-4o",
                # model="gpt-4o-mini",
                messages=[
                    # {"role": "system", "content": "Assistant is a large language model trained by OpenAI."},
                    {"role": "user", "content": prompt}
                ],
                # stop=["\nObservation:"],
                temperature=0.01,
                timeout=40
            )

        generated_message = response.choices[0].message
        text = generated_message.content

        ori_generated_res.append(text)

        sql = parser_sql(text)
        generated_res.append(sql)
        
        if i % 50 == 0:
            generate_sql_file(sql_lst=generated_res, output_path=output_name)
        
            generate_sql_file(sql_lst=ori_generated_res, output_path=unparser_name)   # 未解析的结果保存

    return generated_res, ori_generated_res


def parser_sql(text):
    text = text.strip()
    sql_query_1 = re.search(r'```sql(.*?)```', text, re.DOTALL)
    sql_query_2 = re.search(r'```(.*?)```', text, re.DOTALL)
    if sql_query_1:
        extracted_sql = sql_query_1.group(1).strip()
    elif sql_query_2:
        extracted_sql = sql_query_2.group(1).strip()
    else:
        top_word = text.split(" ")[0]
        if not top_word.lower().startswith("select"):
            extracted_sql = "SELECT " + text
        else:
            extracted_sql = text
    extracted_sql_ls = extracted_sql.split("\n")
    extracted_sql_ls = [s for s in extracted_sql_ls if not s.lower().startswith("-- ") ]
    extracted_sql = "\n".join(extracted_sql_ls)
    return extracted_sql

def collect_response_from_gpt(model_path, gpus_num, db_path_list, question_list, knowledge_list=None, args=None):
    '''
    :param db_path: str
    :param question_list: []
    :return: dict of responses collected from llm
    '''
    responses_dict = {}
    response_list = []

    prompt_ls = []
    for i in tqdm(range(len(question_list)), desc="get prompt"):
        # print('--------------------- processing {}th question ---------------------'.format(i))
        # print('the question is: {}'.format(question))
        question = question_list[i]
        if knowledge_list:
            cur_prompt = generate_combined_prompts_one(db_path=db_path_list[i], question=question, knowledge=knowledge_list[i])
        else:
            cur_prompt = generate_combined_prompts_one(db_path=db_path_list[i], question=question)
        prompt_ls.append(cur_prompt)
    
    if args.use_gpt_api:
        outputs_sql, ori_outputs_text = gpt_generate_result(model_path, gpus_num, prompt_ls, args)
    else:
        outputs_sql, ori_outputs_text = llm_generate_result(model_path, gpus_num, prompt_ls, args)
    
    for i in tqdm(range(len(question_list)), desc="postprocess result"):
        question = question_list[i]
        sql = outputs_sql[i]
        
        db_id = db_path_list[i].split('/')[-1].split('.sqlite')[0]
        sql = sql + '\t----- bird -----\t' + db_id # to avoid unpredicted \t appearing in codex results
        response_list.append(sql)

    return response_list, ori_outputs_text

def question_package(data_json, knowledge=False):
    question_list = []
    for data in data_json:
        question_list.append(data['question'])

    return question_list

def knowledge_package(data_json, knowledge=False):
    knowledge_list = []
    for data in data_json:
        knowledge_list.append(data['evidence'])

    return knowledge_list

def decouple_question_schema(datasets, db_root_path):
    question_list = []
    db_path_list = []
    knowledge_list = []
    for i, data in enumerate(datasets):
        question_list.append(data['question'])
        cur_db_path = os.path.join(db_root_path, data['db_id'], f"{data['db_id']}.sqlite")
        db_path_list.append(cur_db_path)
        knowledge_list.append(data['evidence'])
    
    return question_list, db_path_list, knowledge_list

def generate_sql_file(sql_lst, output_path=None):
    result = {}
    for i, sql in enumerate(sql_lst):
        result[i] = sql
    
    if output_path:
        directory_path = os.path.dirname(output_path)  
        new_directory(directory_path)
        json.dump(result, open(output_path, 'w'), indent=4)
    
    return result

def generate_main(eval_data, args):

    question_list, db_path_list, knowledge_list = decouple_question_schema(datasets=eval_data, db_root_path=args.db_root_path)
    assert len(question_list) == len(db_path_list) == len(knowledge_list)
    
    if args.use_knowledge == 'True':
        responses, ori_outputs_text = collect_response_from_gpt(model_path=args.model_path, gpus_num=args.gpus_num, db_path_list=db_path_list, question_list=question_list, knowledge_list=knowledge_list, args=args)
    else:
        responses, ori_outputs_text = collect_response_from_gpt(model_path=args.model_path, gpus_num=args.gpus_num, db_path_list=db_path_list, question_list=question_list, knowledge_list=None, args=args)
    
    if args.chain_of_thought == 'True':
        output_name = os.path.join(args.data_output_path, f'{args.eval_data_name}_{args.mode}_cot.json')
    else:
        output_name =  os.path.join(args.data_output_path, f'{args.eval_data_name}_{args.mode}_{args.is_use_knowledge}.json')
        unparser_name = os.path.join(args.data_output_path,f'{args.eval_data_name}_{args.mode}_{args.is_use_knowledge}_unparser.json')

    # pdb.set_trace()
    generate_sql_file(sql_lst=responses, output_path=output_name)
    
    generate_sql_file(sql_lst=ori_outputs_text, output_path=unparser_name)   # 未解析的结果保存

    print('successfully collect results from {} for {} evaluation; Use knowledge: {}; Use COT: {}'.format(args.model_path, args.mode, args.use_knowledge, args.chain_of_thought))
    print(f'output: {output_name}')
    # 返回推理数据保存路径
    return output_name