思路

主要为开发者提供一个思路，这里并不是完整的商业项目，只是一时兴起写的一份demo,希望对大家有帮助。

1. 制作一个接口用于上传文件
2. 写一个程序把文件上传到上面的接口中
3. 对得到的文件进行提取，分析（调gpt)

开源仓库地址：GPT-PDF

接口代码

from flask import Flask, request, Response
import PyPDF2

app = Flask(__name__)


@app.route('/upload', methods=['POST'])
def upload_file():
    if 'pdf' not in request.files:
        return "No file part", 400

    file = request.files['pdf']
    if file.filename == '':
        return "No selected file", 400

    if file:
        try:
            reader = PyPDF2.PdfReader(file)
            num_pages = len(reader.pages)
            text = ''

            for page in range(num_pages):
                page_obj = reader.pages[page]
                text += page_obj.extract_text()

            # 指定返回类型为text/plain和编码为utf-8
            return Response(text, mimetype="text/plain", content_type="text/plain; charset=utf-8")

        except Exception as e:
            return str(e), 500


if __name__ == '__main__':
    app.run(debug=True)

上传代码

# coding=gbk
import requests

url = 'http://localhost:5000/upload'
files = {'pdf': open('2.pdf', 'rb')}
response = requests.post(url, files=files)

# 直接打印文本而不是编码文本
print(response.text)

pdf转文本代码

# coding=gbk
# pip install pypdf2 --upgrade

import PyPDF2

# 打开PDF文件
with open('2.pdf', 'rb') as file:
    reader = PyPDF2.PdfReader(file)

    # 获取PDF的总页数
    num_pages = len(reader.pages)

    # 逐页读取
    for page in range(num_pages):
        page_obj = reader.pages[page]
        print(page_obj.extract_text())

综合上述步骤完整代码

import http.client
import json
import requests
# import time

# 开始计时
# start_time = time.time()

# 获取PDF文本
url = 'http://localhost:5000/upload'
files = {'pdf': open('3.pdf', 'rb')}
response = requests.post(url, files=files)
long_text = response.text  # 从接口获得的长文本


# print(long_text)

# 分段函数
def split_text(text, max_size):
    for start in range(0, len(text), max_size):
        yield text[start:start + max_size]


# 配置GPT API   api.zhangsan.cloud
conn = http.client.HTTPSConnection("api.zhangsan.cloud")
headers = {
    'Accept': 'application/json',
    'Authorization': 'Bearer sk-zkyXXXXXXXXXXXXXXXaA47c77',
    'User-Agent': 'Apifox/1.0.0 (https://apifox.com)',
    'Content-Type': 'application/json'
}

# 准备发送到GPT API的消息
all_responses = []

# 系统提示，加入到第一个消息段
system_prompt = "请总结本篇论文,并详细告诉我论文中是基于什么背景.例如：用到了什么方法/算法，是怎么解决的，得到了什么结果，一步步详细告诉我，reply in chinese."

for i, segment in enumerate(split_text(long_text, 8000)):
    if i == 0:
        # 第一个段落，添加系统提示
        messages = [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": segment}
        ]
    else:
        messages = [
            {"role": "user", "content": segment}
        ]

    payload = json.dumps({
        "model": "gpt-3.5-turbo-16k-0613",
        "messages": messages
    })

    conn.request("POST", "/v1/chat/completions", payload, headers)
    res = conn.getresponse()
    data = res.read()
    all_responses.append(json.loads(data.decode("utf-8")))

# 打印或处理所有的响应
for response in all_responses:
    content = response["choices"][0]["message"]["content"]
    print(content)


# print('\n\n')
# # 结束计时并输出运行时间
# end_time = time.time()
# print("Flask API 请求运行时间: {:.2f}秒".format(end_time - start_time))

效果

先运行接口，在运行分析。

效果如下：