#AI夏令营 #Datawhale #夏令营从零入门 AI 逻辑推理-Task 1

Task1 的内容如下：

题目如下：

工具/技术：魔搭Notebook+阿里QWEN大模型API

主要考察点：prompt+fineturning，Task1主要是搭建环境，熟悉流程

主要代码说明：

• 1、improt 所需的库
• 2、日志重新定向到本地文件logs/

from multiprocessing import Process, Manager
import json
import os
from pprint import pprint
import re
from tqdm import tqdm
import random

import uuid
import openai
import tiktoken
import json
import numpy as np
import requests
from retry import retry
from scipy import sparse
#from rank_bm25 import BM25Okapi
#import jieba
from http import HTTPStatus
import dashscope


from concurrent.futures import ThreadPoolExecutor, as_completed
from loguru import logger
import json
import time
from tqdm import tqdm

logger.remove()  # 移除默认的控制台输出
logger.add("logs/app_{time:YYYY-MM-DD}.log", level="INFO", rotation="00:00", retention="10 days", compression="zip")

MODEL_NAME = 'qwen2-7b-instruct'

•  通过dashscope API调用qwen，角色：用户；内容：query

def call_qwen_api(MODEL_NAME, query):
    # 这里采用dashscope的api调用模型推理，通过http传输的json封装返回结果
    messages = [
        {'role': 'user', 'content': query}]
    response = dashscope.Generation.call(
        MODEL_NAME,
        messages=messages,
        result_format='message',  # set the result is message format.
    )
    if response.status_code == HTTPStatus.OK:
        # print(response)
        return response['output']['choices'][0]['message']['content']
    else:
        print('Request id: %s, Status code: %s, error code: %s, error message: %s' % (
            response.request_id, response.status_code,
            response.code, response.message
        ))
        raise Exception()

• 定义提示模版函数 

# 这里定义了prompt推理模版

def get_prompt(problem, question, options):

    options = '\n'.join(f"{'ABCDEFG'[i]}. {o}" for i, o in enumerate(options))

    prompt = f"""你是一个逻辑推理专家，擅长解决逻辑推理问题。以下是一个逻辑推理的题目，形式为单项选择题。所有的问题都是（close-world assumption）闭世界假设，即未观测事实都为假。请逐步分析问题并在最后一行输出答案，最后一行的格式为"答案是：A"。题目如下：

### 题目:
{problem}

### 问题:
{question}
{options}
"""
    # print(prompt)
    return prompt

• 起线程池调用QWEN API，并包装获取的结果

def process_datas(datas,MODEL_NAME):
    results = []
    with ThreadPoolExecutor(max_workers=16) as executor:
        future_data = {}
        lasttask = ''
        lastmark = 0
        lens = 0
        for data in tqdm(datas, desc="Submitting tasks", total=len(datas)):
            problem = data['problem']
            for id,question in enumerate(data['questions']):
                prompt = get_prompt(problem, 
                                    question['question'], 
                                    question['options'],
                                    )

                future = executor.submit(api_retry, MODEL_NAME, prompt)
                
                future_data[future] = (data,id)
                time.sleep(0.6)  # 控制每0.5秒提交一个任务
                lens += 1
        for future in tqdm(as_completed(future_data), total=lens, desc="Processing tasks"):
            # print('data',data)
            data = future_data[future][0]
            problem_id = future_data[future][1]
            try:
                res  = future.result()
                extract_response = extract(res)
                # print('res',extract_response)
                data['questions'][problem_id]['answer'] = extract_response
                results.append(data)
                # print('data',data)
                
            except Exception as e:
                logger.error(f"Failed to process text: {data}. Error: {e}")
    
    return results

其他代码，主要做数据集的输入和输出的处理和校验，就不详细说明了

最后跑完提交文件，拿下第一个分数！(点击即可跳转官网进行提交)

最后贴上我的得分：