'''
1. 基于ChatGPT的多场景应用：
1. 核心模式
1. 联网模式
1. 知识库模式
1. 数据分析模式
1. 智能体模式
1. RAG：
1. 核心文件包括：
1. langchain_KB.py包含了形成vector database的函数，和产生total prompt的函数。
1. rag_source.py包含了从vector database中提取信息来源的函数，包括文档名称和页码。
'''
# TODO：1. 更新huggingface上code01的版本，包括：知识库和数据分析模块。 2. 将知识库模块更新为：multi-query + source。 3. 将数据分析模块重写。
import numpy as np
import pandas as pd
from dotenv import load_dotenv # pip3 install python-dotenv
import requests
from codeinterpreterapi import CodeInterpreterSession, File
import streamlit as st
import openai
import os
import matplotlib.pyplot as plt
import xlrd
import pandas as pd
# import csv
import tempfile
from tempfile import NamedTemporaryFile
import pathlib
from pathlib import Path
from matplotlib.font_manager import FontProperties
import seaborn as sns
from time import sleep
import streamlit_authenticator as stauth
# from langchain.chat_models import ChatOpenAI
from langchain.llms import openai
import sys
import time
import PyPDF2 ## read the local_KB PDF file.
# import localKB_construct
# from streamlit_option_menu import option_menu
# import st_reset_conversation
from st_reset_conversation import reset_all, reset_message
import save_database_info
import pytz
from datetime import datetime
from dotenv import load_dotenv
from openai import OpenAI
import st_msautogen
import rag_source
import add_fonts
import asyncio
import warnings
warnings.filterwarnings("ignore")
#make it look nice from the start
# st.set_page_config(layout='wide',initial_sidebar_state='collapsed',)
### 设置openai的API key
load_dotenv()
openai.api_key = os.environ['user_token']
os.environ["OPENAI_API_KEY"] = os.environ['user_token']
bing_search_api_key = os.environ['bing_api_key']
# # #* 如果数据分析模块在本地调试时碰到接口问题，可以启用如下设置。还可能是一个bash命令的问题，见ChatGPT讲课要点.txt.
openai.proxy = {
"http": "http://127.0.0.1:7890",
"https": "http://127.0.0.1:7890"
}
## layout settings.
st.title("专业版大语言模型智能中心")
st.subheader("Artificial Intelligence Backend Center for Professionals")
st.caption("_声明：本网站仅提供技术测试与评估服务。内容由人工智能生成，仅供参考。如果您本人使用或对外传播本服务生成的输出，您应当主动核查输出内容的真实性、准确性，避免传播虚假信息。_")
# st.divider()
# ## clear conversion.
# def reset_all():
# # st.session_state.conversation = None
# st.session_state.chat_history = None
# st.session_state.messages = []
# message_placeholder = st.empty()
# return None
# navigation menu using Hydralit. 并没有解决menu跳转的问题。
# option_data = [
# {'icon': "house", 'label':"核心模式"},
# {'icon':"cloud-upload",'label':"信息检索模式"},
# {'icon': "gear", 'label':"数据分析模式"},
# {'icon': "list-task", 'label':"智能体模式"},
# ]
# navi_menu = op = hc.option_bar(option_definition=option_data,title=None,key='PrimaryOption', horizontal_orientation=True)
# navi_menu = hc.nav_bar(menu_definition=option_data, key='navi_menu', use_animation=True, option_menu=False, sticky_mode='pinned', sticky_nav=False, hide_streamlit_markers=False)
### 使用streamlit_option_menu格式的类似横幅选项。但是会出现第一次无法运行，需要手动清零或者做一个动作，才可以。
# navi_menu = option_menu(
# menu_title=None,
# options=['核心模式', '信息检索模式', '数据分析模式', '智能体模式'],
# # options=['GPT-3.5', 'GPT-4.0','清华GLM2-6B','百川Baichuan-13B', '阿里通义千问14B'],
# icons=['house', 'cloud-upload','gear','list-task'],
# menu_icon='cast',
# default_index=0,
# orientation='horizontal',
# # manual_select=0,
# # styles={
# # "container": {"padding": "0!important", "background-color": "#fafafa"},
# # "icon": {"color": "orange", "font-size": "25px"},
# # "nav-link": {"font-size": "25px", "text-align": "left", "margin":"0px", "--hover-color": "#eee"},
# # "nav-link-selected": {"background-color": "green"},
# # }
# )
### 常规streamlit选择
navi_menu = st.radio(label='选择一个大语言模型工作模式', options=['核心模式', '联网模式', '知识库模式','数据分析模式', '智能体模式'],index=0,horizontal=True)
# navi_menu = st.selectbox('选择一个大语言模型工作模式', ['核心模式', '信息检索模式', '数据分析模式', '智能体模式'],index=0) ### 原始agent001模式。
reset_button_key = "reset_button"
reset_button = st.button(label=("清除所有记录，并开启一轮新对话 ▶"),
key=reset_button_key, use_container_width=True, type="primary")
def clear_all():
st.session_state.conversation = None
st.session_state.chat_history = None
st.session_state.messages = []
message_placeholder = st.empty()
return None
## 清除所有对话记录, reset all conversation.
if reset_button:
reset_all()
### 上传文件的模块
def upload_file(uploaded_file):
if uploaded_file is not None:
# filename = uploaded_file.name
# st.write(filename) # print out the whole file name to validate. not to show in the final version.
try:
# if '.pdf' in filename: ### original code here.
if '.pdf' in uploaded_file.name:
pdf_filename = uploaded_file.name ### original code here.
filename = uploaded_file.name
# print('PDF file:', pdf_filename)
# with st.status('正在为您解析新知识库...', expanded=False, state='running') as status:
spinner = st.spinner('正在为您解析新知识库...请耐心等待')
with spinner:
### 一下是llama_index方法，但是升级后，可能会报错。
# import localKB_construct
# # st.write(upload_file)
# localKB_construct.process_file(uploaded_file, username)
# ## 在屏幕上展示当前知识库的信息，包括名字和加载日期。
# save_database_info.save_database_info(f'./{username}/database_name.csv', filename, str(datetime.now(pytz.timezone('Asia/Shanghai')).strftime("%Y-%m-%d %H:%M")))
# st.markdown('新知识库解析成功，请务必刷新页面，然后开启对话 🔃')
### 以下是langchain方案。
import langchain_KB
import save_database_info
uploaded_file_name = "File_provided"
temp_dir = tempfile.TemporaryDirectory()
# ! working.
uploaded_file_path = pathlib.Path(temp_dir.name) / uploaded_file_name
with open(pdf_filename, 'wb') as output_temporary_file:
# with open(f'./{username}_upload.pdf', 'wb') as output_temporary_file: ### original code here. 可能会造成在引用信息来源时文件名不对的问题。
# ! 必须用这种格式读入内容，然后才可以写入temporary文件夹中。
# output_temporary_file.write(uploaded_file.getvalue())
output_temporary_file.write(uploaded_file.getvalue())
langchain_KB.langchain_localKB_construct(output_temporary_file, username)
## 在屏幕上展示当前知识库的信息，包括名字和加载日期。
save_database_info.save_database_info(f'./{username}/database_name.csv', pdf_filename, str(datetime.now(pytz.timezone('Asia/Shanghai')).strftime("%Y-%m-%d %H:%M")))
st.markdown('新知识库解析成功，请务必刷新页面，然后开启对话 🔃')
return pdf_filename
else:
# if '.csv' in filename: ### original code here.
if '.csv' in uploaded_file.name:
print('start the csv file processing...')
csv_filename = uploaded_file.name
filename = uploaded_file.name
csv_file = pd.read_csv(uploaded_file)
csv_file.to_csv(f'./{username}/{username}_upload.csv', encoding='utf-8', index=False)
st.write(csv_file[:3]) # 这里只是显示文件，后面需要定位文件所在的绝对路径。
else:
xls_file = pd.read_excel(uploaded_file)
xls_file.to_csv(f'./{username}_upload.csv', index=False)
st.write(xls_file[:3])
print('end the csv file processing...')
# uploaded_file_name = "File_provided"
# temp_dir = tempfile.TemporaryDirectory()
# ! working.
# uploaded_file_path = pathlib.Path(temp_dir.name) / uploaded_file_name
# with open('./upload.csv', 'wb') as output_temporary_file:
# with open(f'./{username}_upload.csv', 'wb') as output_temporary_file:
# print(f'./{name}_upload.csv')
# ! 必须用这种格式读入内容，然后才可以写入temporary文件夹中。
# output_temporary_file.write(uploaded_file.getvalue())
# st.write(uploaded_file_path) #* 可以查看文件是否真实存在，然后是否可以
except Exception as e:
st.write(e)
## 以下代码是为了解决上传文件后，文件路径和文件名不对的问题。
# uploaded_file_name = "File_provided"
# temp_dir = tempfile.TemporaryDirectory()
# # ! working.
# uploaded_file_path = pathlib.Path(temp_dir.name) / uploaded_file_name
# # with open('./upload.csv', 'wb') as output_temporary_file:
# with open(f'./{name}_upload.csv', 'wb') as output_temporary_file:
# # print(f'./{name}_upload.csv')
# # ! 必须用这种格式读入内容，然后才可以写入temporary文件夹中。
# # output_temporary_file.write(uploaded_file.getvalue())
# output_temporary_file.write(uploaded_file.getvalue())
# # st.write(uploaded_file_path) # * 可以查看文件是否真实存在，然后是否可以
# # st.write('Now file saved successfully.')
# return pdf_filename, csv_filename
return filename
### 互联网搜索模块
bing_search_api_key = os.environ['bing_api_key']
bing_search_endpoint = 'https://api.bing.microsoft.com/v7.0/search'
def search(query):
# Construct a request
# mkt = 'en-EN'
mkt = 'zh-CN'
params = {'q': query, 'mkt': mkt}
headers = {'Ocp-Apim-Subscription-Key': bing_search_api_key}
# Call the API
try:
response = requests.get(bing_search_endpoint,
headers=headers, params=params)
response.raise_for_status()
json = response.json()
return json["webPages"]["value"]
# print("\nJSON Response:\n")
# pprint(response.json())
except Exception as e:
raise e
# async def text_mode():
def text_mode():
reset_message() ## reset the message and placeholder.
print('text mode starts!')
# Set a default model
if "openai_model" not in st.session_state:
st.session_state["openai_model"] = "gpt-3.5-turbo-16k"
if radio_1 == 'ChatGPT-3.5':
# print('----------'*5)
print('radio_1: GPT-3.5 starts!')
st.session_state["openai_model"] = "gpt-3.5-turbo-16k"
elif radio_1 == 'ChatGPT-4':
print('radio_1: GPT-4.0 starts!')
st.session_state["openai_model"] = "gpt-4-1106-preview"
else:
st.markdown("**当前大模型无效，请在左侧工具栏中选择一个有效的模型。您现在访问的站点仅提供ChatGPT中的GPT-3.5/4。**")
print(st.session_state["openai_model"])
# Initialize chat history
if "messages" not in st.session_state:
st.session_state.messages = []
# Display chat messages from history on app rerun
for message in st.session_state.messages:
with st.chat_message(message["role"]):
st.markdown(message["content"])
# Display assistant response in chat message container
# if prompt := st.chat_input("说点什么吧"):
prompt = st.chat_input("说点什么吧...")
print('prompt now:', prompt)
print('----------'*5)
if prompt:
st.session_state.messages.append({"role": "user", "content": prompt})
with st.chat_message("user"):
st.markdown(prompt)
with st.chat_message("assistant"):
message_placeholder = st.empty()
full_response = ""
if navi_menu == '联网模式':
# if (navi_menu=='信息检索模式') and (radio_2=='互联网'): ### original code here.
print('联网模式下的prompt:', prompt)
input_message = prompt
internet_search_result = search(input_message)
search_prompt = [
f"Source:\nTitle: {result['name']}\nURL: {result['url']}\nContent: {result['snippet']}" for result in internet_search_result]
prompt = "基于如下的互联网公开信息, 回答问题:\n\n" + \
"\n\n".join(search_prompt[:3]) + "\n\n问题: " + input_message + \
"你需要注意的是回答问题时必须用提问的语言（如英文或者中文）来提示：'答案基于互联网公开信息。'" + "\n\n答案: " # 限制了只有3个搜索结果。
# prompt = "Use these sources to answer the question:\n\n" + "\n\n".join(search_prompt[0:3]) + "\n\nQuestion: " + input_message + "（注意：回答问题时请提示'以下答案基于互联网公开信息。'）\n\n" + "\n\nAnswer: "
st.session_state.messages.append(
{"role": "user", "content": prompt})
## old version of openai API.
# for response in openai.ChatCompletion.create(
# model=st.session_state["openai_model"],
# messages=[
# {"role": m["role"], "content": m["content"]}
# for m in st.session_state.messages
# ],
# stream=True,
# ):
# full_response += response.choices[0].delta.get(
# "content", "")
# message_placeholder.markdown(full_response + "▌")
# message_placeholder.markdown(full_response)
# st.session_state.messages.append(
# {"role": "assistant", "content": full_response})
# st.session_state.messages = []
## new version of openai API.
openai_client = OpenAI()
for response in openai_client.chat.completions.create(
model=st.session_state["openai_model"],
messages=[
{"role": m["role"], "content": m["content"]}
for m in st.session_state.messages
],
stream=True,
):
if str(response.choices[0].delta.content) != 'None':
full_response += str(response.choices[0].delta.content)
message_placeholder.markdown(full_response + "▌")
message_placeholder.markdown(full_response)
st.session_state.messages.append(
{"role": "assistant", "content": full_response})
st.session_state.messages = []
# elif radio_2 != '互联网':
else:
print('ChatGPT only starts!!!')
## 这里需要确认是直接从import openai中获得的函数，而不是langchain中调用openai，否则随着langchain的更新，会出现问题。
# for response in openai.ChatCompletion.create(
# model=st.session_state["openai_model"],
# max_tokens=max_tokens,
# temperature=temperature,
# top_p=top_p,
# presence_penalty=presence_penalty,
# frequency_penalty=frequency_penalty,
# ## 多轮会话，需要记住历史记录。
# messages=[
# {"role": m["role"], "content": m["content"]}
# for m in st.session_state.messages
# ],
# # messages=[{'role': 'system', 'content': 'you are ChatGPT'}, {
# # 'role': 'user', 'content': prompt}], ## 这是单轮会话。
# stream=True,
# ):
openai_client = OpenAI()
for response in openai_client.chat.completions.create(
model=st.session_state["openai_model"],
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
presence_penalty=presence_penalty,
frequency_penalty=frequency_penalty,
## 多轮会话，需要记住历史记录。
messages=[
{"role": m["role"], "content": m["content"]}
for m in st.session_state.messages
],
# messages=[{'role': 'system', 'content': 'you are ChatGPT'}, {
# 'role': 'user', 'content': prompt}], ## 这是单轮会话。
stream=True,
):
# print('full response now:', full_response)
# print('response now:', response)
## old version output format.
# full_response += response.choices[0].delta.get(
# "content", "")
## new version output format.
if str(response.choices[0].delta.content) != 'None': ## 注意这里是内容，而不是response，否则一个chunk的回复。
# print('response now:',response)
full_response += str(response.choices[0].delta.content)
message_placeholder.markdown(full_response + "▌")
message_placeholder.markdown(full_response)
st.session_state.messages.append(
{"role": "assistant", "content": full_response})
## load the local_KB PDF file.
# # def local_KB(uploaded_file):
# print('now starts the local KB version of ChatGPT')
# max_input_size = 4096
# # set number of output tokens
# # num_outputs = 3000 #* working
# num_outputs = 1000
# # set maximum chunk overlap
# max_chunk_overlap = -1000 #* working
# # set chunk size limit
# # chunk_size_limit = 600
# chunk_size_limit = 6000 #* working
# history = []
# if input:
# # ! 这里需要重新装载一下storage_context。
# QA_PROMPT_TMPL = (
# "We have provided context information below. \n"
# "---------------------\n"
# "{context_str}"
# "\n---------------------\n"
# "Given all this information, please answer the following questions,"
# "You MUST use the SAME language as the question:\n"
# "{query_str}\n")
# QA_PROMPT = QuestionAnswerPrompt(QA_PROMPT_TMPL)
# llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0.8, model_name="gpt-3.5-turbo", max_tokens=4096,streaming=True))
# prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)
# service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
# # # index = load_index_from_storage(storage_context)
# storage_context = StorageContext.from_defaults(persist_dir="./")
# index = load_index_from_storage(storage_context,service_context=service_context)
# # query_engine = index.as_query_engine(streaming=True, similarity_top_k=3, text_qa_template=QA_PROMPT)
# # query_engine = index.as_query_engine(streaming=True)
# query_engine = index.as_query_engine(streaming=True, text_qa_template=QA_PROMPT)
# reply = query_engine.query(input)
# async def localKB_mode(username):
def localKB_mode(username):
# reset_all() ## reset the conversation.
reset_message() ## only reset the message and placeholder.
print('now starts the local KB version of ChatGPT')
# # Initialize chat history
# if "messages" not in st.session_state:
# st.session_state.messages = []
# for message in st.session_state.messages:
# with st.chat_message(message["role"]):
# st.markdown(message["content"])
# Display assistant response in chat message container
# if prompt:
if prompt := st.chat_input("说点什么吧"):
st.session_state.messages.append({"role": "user", "content": prompt})
with st.chat_message("user"):
st.markdown(prompt)
with st.status('检索中...', expanded=True, state='running') as status:
# try:
with st.chat_message("assistant"):
message_placeholder = st.empty()
full_response = ""
### llama_index框架的RAG代码，最近更新版本后不成功，会报错。
### outdated version.
# llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0.8, model_name="gpt-3.5-turbo", max_tokens=4024,streaming=True))
# # print('llm_predictor:', llm_predictor)
# prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)
# print('prompt_helper:', prompt_helper)
# service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
# print('service_context:', service_context)
# # # index = load_index_from_storage(storage_context)
# print("storage_context:", storage_context)
# index = load_index_from_storage(storage_context,service_context=service_context)
## sample code for reference.
# docstore = 'storage/docstore.json'
# index_store = 'storage/index_store.json'
# vector_store = 'storage/vector_store.json'
# print('storage_context:', storage_context)
##NOTE: 这里需要重新装载一下storage_context。
# storage_context = StorageContext.from_defaults(persist_dir=f"./{username}/")
# print('--'*30)
# print('storage_context:',storage_context)
# print('type of storage_context.index_store:', type(storage_context.index_store))
# print('--'*30)
# # storage_context = {storage_context}
# index = load_index_from_storage(storage_context)
# print('--'*30)
# print('index now:', index)
# context_str = index
# ##TODO 重新构建Prompt，加入QA_Template.
# QA_PROMPT_TMPL = (
# "We have provided context information below. \n"
# "---------------------\n"
# "{context_str}"
# "\n---------------------\n"
# "Given all this information, please answer the following questions,"
# "You MUST use the SAME language as the question and the default language is Chinese:\n"
# "{query_str}\n")
# # QA_PROMPT = QuestionAnswerPrompt(QA_PROMPT_TMPL) ## outdated version.
# ##TODO: newer version but may run into llama_index import problem.
# # qa_template = PromptTemplate(QA_PROMPT_TMPL)
# # prompt = qa_template.format(context_str=context_str, query_str=prompt)
# # prompt = qa_template.format(context_str=context_str, query_str=QA_PROMPT)
# # query_engine = index.as_query_engine(streaming=True, similarity_top_k=3, text_qa_template=QA_PROMPT)
# query_engine = index.as_query_engine(streaming=False)
# print('111')
## older version.
# query_engine = index.as_query_engine(streaming=True, text_qa_template=QA_PROMPT)
# query_engine = index.as_query_engine()
# reply = query_engine.query(prompt)
# llama_index_reply = query_engine.query(prompt)
# # full_response += query_engine.query(prompt)
# print('local KB reply:', llama_index_reply)
# # query_engine.query(prompt).print_response_stream() #* 能在terminal中流式输出。
# # for resp in llama_index_reply.response_gen:
# # print(resp)
# # full_response += resp
# # message_placeholder.markdown(full_response + "▌")
# message_placeholder.markdown(str(llama_index_reply))
# print('333')
# # st.session_state.messages.append(
# # {"role": "assistant", "content": full_response})
# # st.session_state.messages = []
# # full_response += reply
# # full_response = reply
# # st.session_state.messages.append(
# # {"role": "assistant", "content": full_response})
### 用langchain的FAISS来做RAG
import langchain_KB
### 默认选择使用multi-query的方法进行查询。
##NOTE: 目前这个版本是把所有的multiquery当成一个问题提交给大模型。后续可以考虑将每一个问题分别送入大模型，然后得到的多个答案，然后在一并汇总。
if True:
import multiQuery_prompt
prompt = multiQuery_prompt.generate_sim_query(orignal_question=prompt)
# print('multiQuery prompts now:', prompt)
prompt = str(prompt) ## 需要强制转换成str格式。
total_prompt, docs = langchain_KB.langchain_RAG(prompt, username)
print('total_prompt now:', total_prompt)
st.session_state.messages.append({"role": "user", "content": total_prompt})
## new version of openai API.
openai_client = OpenAI()
for response in openai_client.chat.completions.create(
model=st.session_state["openai_model"],
messages=[
{"role": m["role"], "content": m["content"]}
for m in st.session_state.messages
],
stream=True,
):
if str(response.choices[0].delta.content) != 'None':
full_response += str(response.choices[0].delta.content)
message_placeholder.markdown(full_response + "▌")
message_placeholder.markdown(full_response)
st.session_state.messages.append(
{"role": "assistant", "content": full_response})
st.session_state.messages = []
try:
### 显示RAG的source，即查询得到的信息来源出处。
print('docs now:', docs)
source = rag_source.rag_source(docs) ## get the k reference source of the RAG answer, in a designed format.
# print('返回的source内容：', source)
st.divider()
st.caption(source)
except Exception as e:
print('Exception:', e)
pass
##TODO 确认是否需要？
st.session_state.messages = []
# except Exception as e:
# print('Exception:', e)
# pass
# async def data_mode():
def data_mode():
clear_all() ## reset the conversation.
print('数据分析模式启动！')
# uploaded_file_path = './upload.csv'
uploaded_file_path = f'./{username}/{username}_upload.csv'
# uploaded_file_path = f'./{username}_upload.csv' ### original code here.
print('file path:', uploaded_file_path)
# Initialize chat history
if "messages" not in st.session_state:
st.session_state.messages = []
# Display chat messages from history on app rerun
for message in st.session_state.messages:
with st.chat_message(message["role"]):
st.markdown(message["content"])
# Display assistant response in chat message container
# if prompt := st.chat_input("Say something"):
prompt = st.chat_input("Say something")
print('prompt now:', prompt)
print('----------'*5)
if prompt:
try:
st.session_state.messages.append({"role": "user", "content": prompt})
with st.chat_message("user"):
st.markdown(prompt)
with st.status('思考中...需要1至10分钟左右，请耐心等待 🏃', expanded=True, state='running') as status:
with st.chat_message("assistant"):
#### Using the open-source CodeInterpreter solution below. May not work after version update, need to upgrade the code accoridngly later on.
# from langchain.chat_models import ChatOpenAI
# llm_model = ChatOpenAI(model_name="gpt-4-1106-preview")
# # llm_model = ChatOpenAI(model_name="gpt-4")
# # async with CodeInterpreterSession(llm=llm_model) as session:
# import interpreter
# interpreter.llm.model = "gpt-3.5-turbo"
# with CodeInterpreterSession(llm=llm_model) as session:
# # with CodeInterpreterSession(llm=llm_model) as session:
# print('222')
# # user_request = "对于文件中的'SepalLengthCm’数据给我一个'直方图'，提供图表，并给出分析结果"
# #! 可以用设定dpi=300来输出高质量的图表。（注：图的解析度dpi设定为300）
# environ_settings = """【背景要求】如果我没有告诉你任何定制化的要求，那么请你按照以下的默认要求来回答：
# -------------------------------------------------------------------------
# 1. 你需要用我提问的语言来回答，且默认情况下用中文来回答。
# 2. 如果要求你输出图表，那么图的解析度dpi需要设定为600。图尽量使用seaborn库。seaborn库的参数设定：sns.set(rc={'axes.facecolor':'#FFF9ED','figure.facecolor':'#FFF9ED'}, palette='dark'。
# 3. 图上所有的文字全部翻译成<英文English>来表示。
# 4. 你回答的文字内容必须尽可能的详细且通俗易懂。
# 5. 回答时尽可能地展示分析所对应的图表，并提供分析结果。 你需要按如下格式提供内容：
# 5.1 提供详细且专业的分析结果，提供足够的分析依据。
# 5.2 给出可能造成这一结果的可能原因有哪些？
# 以上内容全部用【1, 2, 3这样的序列号格式】来表达。
# -------------------------------------------------------------------------
# """ # seaborn中的palette参数可以设定图表的颜色,选项包括：deep, muted, pastel, bright, dark, colorblind，Spectral。更多参数可以参考：https://seaborn.pydata.org/generated/seaborn.color_palette.html。
# # uploaded_file_path = upload_file()
# user_request = environ_settings + "\n\n" + \
# "你需要完成以下任务：\n\n" + prompt + "\n\n" \
# f"注：文件位置在 {uploaded_file_path}"
# # user_request = str(prompt) ### only prompt without environment prompt.
# print('user_request: \n', user_request)
# # 加载上传的文件，主要路径在上面代码中。
# files = [File.from_path(str(uploaded_file_path))]
# # files = [File.from_path("/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/rawdata/short_csv.csv")]
# # st.write(pd.DataFrame(files))
# # print('session.__init__', session.__init__)
# # print('session', session.__init__)
# with st.status('思考中...', expanded=True, state='running') as status:
# # generate the response
# # response = await session.generate_response(user_msg=user_request, files=files, detailed_error=True)
# # response = await session.generate_response(user_msg=user_request, files=files)
# response = session.generate_response(user_msg=user_request, files=files)
# # output to the user
# print("AI: ", response.content)
# full_response = response.content
# ### full_response = "this is full response"
# # for file in response.files:
# for i, file in enumerate(response.files):
# # await file.asave(f"/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/output{i}.png") ##working.
# # st.image(file.get_image()) #! working.
# # * 注意这里的设定，可以提高图片的精细程度。
# st.image(file.get_image(), width=None,
# output_format='PNG')
# # message_placeholder.markdown(full_response + "▌") ## orignal code.
# # message_placeholder.markdown(full_response) ## orignal code.
# st.write(full_response)
# status.update(label='complete', state='complete')
# # TODO: 确认是否要记录所有的full response。
# st.session_state.messages.append(
# {"role": "assistant", "content": full_response})
# # await session.astop() # ! 确认需要关闭。
# session.astop() # ! 确认需要关闭。
# # st.session_state.messages.append({"role": "assistant", "content": full_response})
#### #### Using the OpenAI's assistant API， wrap into the st_openai_assistant.py.
import st_openai_assistant
### NOTE:在st_openai_assistant.py中可以设置system_prompt.
# sys_prompt = """ 1. 你是一位智能AI助手，你连接着一台电脑，但请注意不能联网。在使用Python解决任务时，你可以运行代码并得到结果，如果运行结果有错误，你需要尽可能对代码进行改进。你可以处理用户上传到电脑上的文件。
# 2. 你使用matplotlib.pylab（plt）或者seaborn（sns）画图时，需要添加中文字库，代码如下：
# matplotlib.rcParams['font.sans-serif'] = ['Microsoft YaHei UI']
# sns.set(rc={'axes.facecolor':'#FFF9ED','figure.facecolor':'#FFF9ED'}, palette='dark', font='Microsoft YaHei UI')
# 3. 如果我没有告诉你任何定制化的要求，那么请你按照以下的默认要求来回答：
# 3.1 你回答的文字内容必须尽可能的详细且通俗易懂。
# 3.2 回答时尽可能地展示分析所对应的图表，并提供分析结果。图表上的文字采用中文。你需要按如下格式提供内容：
# * 提供详细且专业的分析结果，提供足够的分析依据。
# * 给出可能造成这一结果的可能原因有哪些？
# 以上内容你用序列号1、2、3这样的格式表达。
# """
# prompt = [
# {"role": "system", "content": sys_prompt},
# {"role": "user", "content": prompt},
# ]
messages, text_response, img_response, image_files, final_answer = st_openai_assistant.openai_assistant(prompt=prompt, filepath=uploaded_file_path, username=username)
# st.image(img_response) ## show one single image.
# st.markdown(text_response) ## show one single text response.
try:
from PIL import Image
print("返回到Agent001程序中的图表个数：", len(image_files))
# plt.imshow(img)
# plt.show()
for img in image_files:
img = Image.open(img) ## image object.
st.image(img, output_format='PNG')
# st.image(f"./{username}/{img_response[i]}", output_format='PNG')
# st.image(f'/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/joeshi/output{i}.png', output_format='PNG')
except:
pass
try:
st.markdown(final_answer) ## all messages are appended together, need to print out one by one?
except:
pass
except Exception as e:
print(e)
pass
### authentication with a local yaml file.
import yaml
from yaml.loader import SafeLoader
with open('./config.yaml') as file:
config = yaml.load(file, Loader=SafeLoader)
authenticator = stauth.Authenticate(
config['credentials'],
config['cookie']['name'],
config['cookie']['key'],
config['cookie']['expiry_days'],
config['preauthorized']
)
# authentication with a remove cloud-based database.
# 导入云端用户数据库。
# DETA_KEY = "c0zegv33efm_4MBTaoQAn76GzUfsZeKV64Uh9qMY3WZb"
# load_dotenv(".env")
# DETA_KEY = os.getenv("DETA_KEY")
# print(DETA_KEY)
# deta = Deta(DETA_KEY)
# mybase is the name of the database in Deta. You can change it to any name you want.
# credentials = {"usernames":{}}
# # credentials = {"users": {}}
# # db = db()
# users = []
# email = []
# passwords = []
# names = []
# for row in db.fetch_all_users():
# users.append(row["username"])
# email.append(row["email"])
# names.append(row["key"])
# passwords.append(row["password"])
# hashed_passwords = stauth.Hasher(passwords).generate()
## 需要严格的按照yaml文件的格式来定义如下几个字段。
# for un, name, pw in zip(users, names, hashed_passwords):
# # user_dict = {"name":name,"password":pw}
# user_dict = {"name": un, "password": pw}
# # credentials["usernames"].update({un:user_dict})
# credentials["usernames"].update({un: user_dict})
# ## sign-up模块，未完成。
# database_table = []
# # print(pd.DataFrame(credentials))
# for i in credentials['usernames'].keys():
# # print("i:",i)
# # print("name",credentials['usernames'][i]['name'])
# # print("password",credentials['usernames'][i]['password'])
# database_table.append([i,credentials['usernames'][i]['name'],credentials['usernames'][i]['password']])
# print("database_table:",database_table)
# authenticator = stauth.Authenticate(
# credentials=credentials, cookie_name="joeshi_gpt", key='abcedefg', cookie_expiry_days=30)
user, authentication_status, username = authenticator.login('用户登录', 'main')
# print("name", name, "username", username)
# ## sign-up widget，未完成。
# try:
# if authenticator.register_user('新用户注册', preauthorization=False):
# # for list in database_table:
# # db.update_user(username=list[0], name=list[1], password=list[2])
# db.update_user(username=list[-1][0], name=list[-1][1], password=list[-1][2])
# # st.success('User registered successfully')
# st.success('注册成功！')
# except Exception as e:
# st.error(e)
if authentication_status:
with st.sidebar:
st.markdown(
"""
<style>
[data-testid="stSidebar"][aria-expanded="true"]{
min-width: 450px;
max-width: 450px;
}
""",
unsafe_allow_html=True,
)
### siderbar的题目。
st.header(f'**欢迎 **{username}** 来到人工智能的世界** ♠')
st.write(f'_Welcome and Hope U Enjoy Staying Here_')
authenticator.logout('登出', 'sidebar')
# reset_button_key = "reset_button"
# reset_button = st.button(label=("清除所有记录，并开启一轮新对话 ▶"),
# key=reset_button_key, use_container_width=True, type="primary")
# ## 清除所有对话记录, reset all conversation.
# if reset_button:
# reset_all()
# st.markdown("#### 大语言模型设置")
# with st.expander(label='**选择一个大语言模型基座**', expanded=True):
radio_1 = st.selectbox(
label='选择一个大语言模型基座 (注：根据站点不同，部分基座不可用)',
options=["ChatGPT-4", "ChatGPT-3.5", "Google Gemini","Claude 3", "清华ChatGLM3-6B", "百川Baichuan-13B", "阿里通义千问-14B", "阿里通义千问-72B", "Llama-2", "Mistral", "Vicuna"],
index=0,
placeholder="大语言模型列表",
)
## 在sidebar上的三个分页显示，用st.tabs实现。
tab_1, tab_2, tab_3, tab_4 = st.tabs(['使用须知', '模型参数', '提示词模板', '系统角色设定'])
# with st.expander(label='**使用须知**', expanded=False):
with tab_1:
# st.markdown("#### 快速上手指南")
# with st.text(body="说明"):
# st.markdown("* 重启一轮新对话时，只需要刷新页面（按Ctrl/Command + R）即可。")
with st.text(body="说明"):
st.markdown("* 为了保护数据与隐私，所有对话均不会被保存，刷新页面立即删除。敬请放心。")
with st.text(body="说明"):
st.markdown("* “GPT-4”回答质量极佳，但速度缓慢，建议适当使用。")
with st.text(body="说明"):
st.markdown("* “信息检索模式”与所有的搜索引擎或者数据库检索方式一样，仅限一轮对话，将不会保持之前的会话记录。")
with st.text(body="说明"):
st.markdown(
"* “数据分析模式”暂时只支持1000个单元格以内的数据分析，单元格中的内容不支持中文数据（表头也尽量不使用中文）。一般运行时间在1至10分钟左右，期间需要保持网络畅通。")
with st.text(body="说明"):
st.markdown("* “数据分析模式”推荐上传csv格式的文件，部分Excel文件容易出现数据不兼容的情况。")
## 大模型参数
# with st.expander(label='**大语言模型参数**', expanded=True):
with tab_2:
max_tokens = st.slider(label='Max_Token（生成结果时最大字数）', min_value=100, max_value=8096, value=4096,step=100)
temperature = st.slider(label='Temperature (温度)', min_value=0.0, max_value=1.0, value=0.8, step=0.1)
top_p = st.slider(label='Top_P (核采样)', min_value=0.0, max_value=1.0, value=0.6, step=0.1)
frequency_penalty = st.slider(label='Frequency Penalty (重复度惩罚因子)', min_value=-2.0, max_value=2.0, value=1.0, step=0.1)
presence_penalty = st.slider(label='Presence Penalty (控制主题的重复度)', min_value=-2.0, max_value=2.0, value=1.0, step=0.1)
## reset password widget
# try:
# if authenticator.reset_password(st.session_state["username"], 'Reset password'):
# st.success('Password modified successfully')
# except Exception as e:
# st.error(e)
# with st.header(body="欢迎"):
# st.markdown("# 欢迎使用大语言模型商业智能中心")
# with st.expander(label=("**重要的使用注意事项**"), expanded=True):
# with st.container():
with tab_3:
# st.markdown("#### Prompt提示词参考资料")
with st.expander(label="**大语言模型基础提示词Prompt示例**", expanded=False):
st.code(
body="继续用中文写一篇关于 [文章主题] 的文章，以下列句子开头：[文章开头］。", language='plaintext')
st.code(body="将以下文字概括为 100 个字，使其易于阅读和理解。避免使用复杂的句子结构或技术术语。",
language='plaintext')
st.code(body="给我出一个迪奥2024春季发布会活动策划。", language='plaintext')
st.code(body="帮我按照正式会议结构写一个会邀：主题是xx手机游戏立项会议。", language='plaintext')
st.code(body="帮我写一个车内健康监测全场景落地的项目计划，用表格。", language='plaintext')
st.code(
body="同时掷两枚质地均匀的骰子，则两枚骰子向上的点数之和为 7 的概率是多少。", language='plaintext')
st.code(body="写一篇产品经理的演讲稿，注意使用以下词汇: 赋能,抓手,中台,闭环,落地,漏斗,沉淀,给到,同步,对齐,对标,迭代,拉通,打通,升级,交付,聚焦,倒逼,复盘,梳理,方案,联动,透传,咬合,洞察,渗透,兜底,解耦,耦合,复用,拆解。", language='plaintext')
with st.expander(label="**数据分析模式的专用提示词Prompt示例**", expanded=False):
# with st.subheader(body="提示词Prompt"):
st.code(body="分析此数据集并绘制一些'有趣的图表'。", language='python')
st.code(
body="对于这个文件中的数据，你需要要找出[X,Y]数据之间的寻找'相关性'。", language='python')
st.code(body="对于这个文件中的[xxx]数据给我一个'整体的分析'。", language='python')
st.code(body="对于[xxx]数据给我一个'直方图'，提供图表，并给出分析结果。", language='python')
st.code(body="对于[xxx]数据给我一个'小提琴图'，并给出分析结果。", language='python')
st.code(
body="对于[X,Y,Z]数据在一个'分布散点图 (stripplot)'，所有的数据在一张图上展现, 并给出分析结果。", language='python')
st.code(body="对于[X，Y]数据，进行'T检验'，你需要展示图表，并给出分析结果。",
language='python')
st.code(body="对于[X，Y]数据给我一个3个类别的'聚类分析'，并给出分析结果。",
language='python')
with tab_4:
st.text_area(label='系统角色设定', value='你是一个人工智能，你需要回答我提出的问题，或者完成我交代的任务。你需要使用我提问的语言（如中文、英文）来回答。', height=200, label_visibility='hidden')
elif authentication_status == False:
st.error('⛔ 用户名或密码错误！')
elif authentication_status == None:
st.warning('⬆ 请先登录！')
if __name__ == "__main__":
import asyncio
try:
match navi_menu:
case "核心模式":
# if navi_menu == "核心模式":
print(f'navi_menu 选择了 {navi_menu}')
reset_all()
# * 也可以用命令执行这个python文件。’streamlit run frontend/app.py‘
# asyncio.run(text_mode())
text_mode()
# elif navi_menu == "信息检索模式":
case "联网模式":
# print(f'navi_menu 选择了 {navi_menu}')
# reset_all()
##TODO 如下设置中的index=None, 可能可以解决了刷屏会调回第一项的问题？好像不会。
# radio_2 = st.radio(label='信息检索源选择:', options=['互联网', '维基百科', '本地文档', '文献库', '企业知识库','知识图谱库'], horizontal=True, label_visibility='visible')
### 横向排列的checkbox选项。也可以实现多项选择的功能。
# col_1, col_2, col_3, col_4, col_5 = st.columns(5)
# rag_1 = col_1.checkbox(label='互联网', label_visibility='visible')
# rag_2 = col_2.checkbox(label='上传文件', label_visibility='visible')
# rag_3 = col_3.checkbox(label='企业知识库', label_visibility='visible')
# rag_4 = col_4.checkbox(label='百科全书', label_visibility='visible')
# rag_5 = col_5.checkbox(label='其他数据源', label_visibility='visible')
if (navi_menu=='联网模式'):
# print(f'radio 选择了 {radio_2}')
# asyncio.run(text_mode())
text_mode()
case "知识库模式":
print(f'navi_menu 选择了 {navi_menu}')
st.session_state.messages = []
# ### llama_index框架的RAG代码，最近更新版本后不成功，会报错。
# path = f'./{username}/vector_store.json'
# if os.path.exists(path):
# print(f'{path} local KB exists')
# database_info = pd.read_csv(f'./{username}/database_name.csv')
# current_database_name = database_info.iloc[-1][0]
# current_database_date = database_info.iloc[-1][1]
# database_claim = f"当前知识库为：{current_database_name}，创建于{current_database_date}。可以开始提问！"
# st.markdown(database_claim)
### Langchain框架的RAG代码。
path = f'./{username}/faiss_index/index.faiss'
if os.path.exists(path):
print(f'{path} local KB exists')
database_info = pd.read_csv(f'./{username}/database_name.csv', encoding='utf-8', header=None) ## 不加encoding的话，中文名字的PDF会报错。
print(database_info)
current_database_name = database_info.iloc[-1][0]
current_database_date = database_info.iloc[-1][1]
database_claim = f"当前知识库为：{current_database_name}，创建于{current_database_date}。可以开始提问！"
st.markdown(database_claim)
try:
uploaded_file = st.file_uploader(
"选择上传一个新知识库", type=(["pdf"]))
# 默认状态下没有上传文件，None，会报错。需要判断。
if uploaded_file is not None:
# uploaded_file_path = upload_file(uploaded_file)
upload_file(uploaded_file)
except Exception as e:
print(e)
pass
try:
## 启动本地知识库模式。
localKB_mode(username)
# asyncio.run(localKB_mode(username))
except Exception as e:
print(e)
pass
# elif navi_menu == "数据分析模式":
case "数据分析模式":
# reset_message()
uploaded_file = st.file_uploader(
"选择一个文件", type=(["csv", "xlsx", "xls"]))
# 默认状态下没有上传文件，None，会报错。需要判断。
if uploaded_file is not None:
# uploaded_file_path = upload_file(uploaded_file) ### original code here.
csv_filename = upload_file(uploaded_file)
# asyncio.run(data_mode())
reset_all()
data_mode()
# elif navi_menu == "智能体模式":
case "智能体模式":
uploaded_file = st.file_uploader(
"选择一个文件", type=(["csv"]))
reset_all()
print('st uploaded_file:',uploaded_file)
# 默认状态下没有上传文件，None，会报错。需要判断。
# if uploaded_file is not None:
if uploaded_file is not None:
uploaded_file_path = upload_file(uploaded_file)
# asyncio.run(data_mode())
else:
uploaded_file_path = None
# st.markdown('**此功能还在内部测试阶段，尚未开放，敬请期待！**')
# reset_message()
print('st_msautogen starts!')
uploaded_file_path = '/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/joeshi_upload.csv'
# asyncio.run(st_msautogen.auto_gen(uploaded_file_path)) ## 好像不需要启动asyncio，也可以正常运行。在msautogen中已经启动了。
st_msautogen.auto_gen(uploaded_file_path) ## 这里不需要使用asyncio.run()，否则会streamlit中会刷新页面？
except Exception as e:
print('Exception Raised:',e)
pass