Dunhuang_GPT / app.py
Paperboxiv's picture
debug
30d0a35
raw
history blame
4.8 kB
# -*- coding: utf-8 -*-
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import Chroma
import openai
import time
import gradio as gr
import os
# 输入 API KEY
os.environ["OPENAI_API_KEY"] = "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"
#读取PDF文件
def doc_read_pdf(file):
# 读取PDF
reader = PdfReader(file)
# reader = PdfReader('.\data\資治通鑑全集_部分1.pdf')
raw_text = ''
for i, page in enumerate(reader.pages):
text = page.extract_text()
if text:
raw_text += text
return raw_text
# 读取txt文件
def doc_read_txt(file):
with open(file, encoding='utf-8') as f:
text = f.read()
return text
#补全
#从开始到调用openai模型前的一些步骤,主要是文件读取和拆解
def doc_split(file):
#分解文本
text_splitter = CharacterTextSplitter(
separator = "\n",
chunk_size = 1200,
chunk_overlap = 100,
length_function = len,
)
# texts = text_splitter.split_text(raw_text)
texts = text_splitter.split_text(doc_read_txt(file))
return texts
#将文本向量化
def doc_vectorize(texts):
embeddings = OpenAIEmbeddings()
docsearch = Chroma.from_texts(texts, embeddings, metadatas=[{"source": str(i)} for i in range(len(texts))])
return docsearch
#文本被拆解储存在数组texts中
# raw_file =
texts = doc_split("./资治通鉴_1_残缺.txt")
docsearch = doc_vectorize((texts))
def openai_reply(word1, word2, word3, temp, file):
words = word1 + "*****" + word2 + "*****" + word3
# 文本相似查找,最终结果是一个列表
docs = docsearch.similarity_search(words)
reference_1 = docs[0].page_content
reference_2 = docs[1].page_content
reference = reference_1 + reference_2
print(words)
request = "请使用文言文帮我补全[" + words + "]"
response = openai.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=[
{"role": "system",
"content": f"""你是一个古汉语与中国历史专家,擅长补全古代文献。在后续的会话中,你需要补全我给你的残缺文本,
这些残缺文本用[]包括,并且其中的几段残缺汉字用“*”来代替着,且数量不明。
请你阅读并且理解下文背景资料,并且补全我待会在会话中给出的残缺文本。如果你在背景资料中找不到相关文字,请根据你对于背景资料和我给出的残缺文本的理解,
使用《资治通鉴》的文言文风格自行补全残缺文本。请注意,不要改动我提供的残缺文本中非“*”的原文,并且一定要用文言文替代“*”!!!
\n背景资料:\n{reference}
"""},
{"role": "user", "content": request},
],
max_tokens=512,
n=1,
stop=None,
temperature=temp
)
print(response.choices[0].message['content'])
shijian = time.strftime("%Y年%m月%d日%H点%M分",time.localtime())
answer = response.choices[0].message['content']
return answer, reference, shijian
# 以下是界面搭建
headline = '碎片化文本复原'
description = """请给出至多三条碎片文本,系统将会根据文献数据库尽可能进行理解和匹配,给出猜想。
当然,你也可以上传自己的TXT文件作为数据来源之一。结果仅供参考和启发。"""
with gr.Blocks() as demo:
gr.Markdown(f'<center><h1>{headline}</h1></center>')
gr.Markdown(description)
with gr.Row():
with gr.Group():
raw_text_1 = gr.Textbox(label='在此输入碎片文本')
raw_text_2 = gr.Textbox(label='在此输入碎片文本')
raw_text_3 = gr.Textbox(label='在此输入碎片文本')
temp = gr.Slider(minimum=0.0, maximum=2.0, value=0.3, label="无序程度(Temperature)")
Title = gr.Textbox(label='在此输入提交的材料的标题(无需加《》)')
file = gr.File(
label='上传你的本地TXT文件', file_types=['.txt']
)
btn = gr.Button(value='提交')
btn.style(full_width=True)
with gr.Group():
shijian = gr.Label(label='生成时时间')
answer = gr.Textbox(label='回答')
reference = gr.Textbox(label='参考材料')
btn.click(
openai_reply,
inputs= [raw_text_1, raw_text_2, raw_text_3, temp, file],
outputs= [answer, reference, shijian],
)
demo.launch()
# if __name__ == "__main__":
# demo.launch(server_port=7860, share=True)