Upload 26 files

app.py

@@ -1,33 +1,28 @@
 '''
-参考： https://github.com/shroominic/codeinterpreter-api
-
-1. 可以存在本地，然后再调出来。 working. 
-1. 可以在临时文件夹中读取文件。
-1. 可以直接在内存中读出图片。
-1. 中文字体成功。
-    from matplotlib.font_manager import FontProperties
-    myfont=FontProperties(fname='/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/rawdata/SimHei.ttf')
-    sns.set_style('whitegrid',{'font.sans-serif':['simhei','Arial']}) 
-1. 解决了account login的问题，主要格式:
-    ## 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})
+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. Chinese display isssue. 2. account system. 3. local enterprise database. 
+# 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 database as db
-# from deta import Deta  # pip3 install deta
+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
-# from codeinterpreterapi import CodeInterpreterSession
 import openai
 import os
 import matplotlib.pyplot as plt
+import xlrd
 import pandas as pd
 # import csv
 import tempfile
@@ -38,43 +33,94 @@ from matplotlib.font_manager import FontProperties
 import seaborn as sns
 from time import sleep
 import streamlit_authenticator as stauth
-# import database as db  # python文件同目录下的.py程序，直接导入。
-# import deta
-from langchain.chat_models import ChatOpenAI
-from llama_index import StorageContext, load_index_from_storage, GPTVectorStoreIndex, LLMPredictor, PromptHelper
-from llama_index import ServiceContext, QuestionAnswerPrompt
+# 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
-from datetime import datetime
 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']
-openai.api_key = os.environ['user_token'] 
-# os.environ["VERBOSE"] = "True"  # 可以看到具体的错误？
+bing_search_api_key = os.environ['bing_api_key']
 
-#* 如果碰到接口问题，可以启用如下设置。
-# openai.proxy = {
-#     "http": "http://127.0.0.1:7890",
-#             "https": "http://127.0.0.1:7890"
-# }
 
+# # #* 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("专业版大语言模型商业智能中心")
+## 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���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。
 
-# clear conversion.
 reset_button_key = "reset_button"
-reset_button = st.button(label=("扫清世间烦恼，清除所有记录，并开启一轮新对话 ▶"),
-                         key=reset_button_key, use_container_width=True, type="secondary")
-if reset_button:
-    st.session_state.conversation = None
-    st.session_state.chat_history = None
-    st.session_state.messages = []
-    message_placeholder = st.empty()
+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
@@ -83,73 +129,104 @@ def clear_all():
     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.')
 
-# # with tab2:
-# 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:
-#                 # pdf_file = PyPDF2.PdfReader(uploaded_file)
-#                 PyPDF2.PdfReader(uploaded_file)
-#                 # st.write(pdf_file.pages[0].extract_text())
-#                 # 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 st.spinner('正在为您解析新知识库...请耐心等待'):
-#                 with spinner:
-#                     import localKB_construct
-#                     # sleep(3)
-#                     # st.write(upload_file)
-#                     localKB_construct.process_file(uploaded_file)
-#                     st.markdown('新知识库解析成功，可以开始对话！')
-#                     spinner = st.empty()
-#                     # sleep(3)
-#                     # display = []
-
-#             else:
-#                 if '.csv' in filename:
-#                     csv_file = pd.read_csv(uploaded_file)
-#                     csv_file.to_csv('./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('./upload.csv', index=False)
-#                     st.write(xls_file[:3])
-
-#                 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.')
-#         except Exception as e:
-#             st.write(e)
-
-#         # 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 None
+        # 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'
@@ -169,19 +246,26 @@ def search(query):
     except Exception as e:
         raise e
 
-# openai.api_key = st.secrets["OPENAI_API_KEY"]
-
-async def text_mode():
+# 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 == 'GPT-3.5':
+
+    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"
-    else:
+    elif radio_1 == 'ChatGPT-4':
         print('radio_1: GPT-4.0 starts!')
-        st.session_state["openai_model"] = "gpt-4"
+        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:
@@ -193,11 +277,10 @@ async def text_mode():
             st.markdown(message["content"])
 
     # Display assistant response in chat message container
-    # if prompt := st.chat_input("Say something"):
-    prompt = st.chat_input("Say something")
+    # 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:
     if prompt:
         st.session_state.messages.append({"role": "user", "content": prompt})
         with st.chat_message("user"):
@@ -207,8 +290,9 @@ async def text_mode():
             message_placeholder = st.empty()
             full_response = ""
 
-            if radio_2 == '联网模式':
-                print('联网模式入口，prompt:', prompt)
+            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 = [
@@ -221,7 +305,26 @@ async def text_mode():
                 st.session_state.messages.append(
                     {"role": "user", "content": prompt})
 
-                for response in openai.ChatCompletion.create(
+                ## 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"]}
@@ -229,85 +332,134 @@ async def text_mode():
                     ],
                     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 = []
-
-            if radio_2 == '核心模式':
-                print('GPT only starts!!!')
-                print('messages:', st.session_state['messages'])
-                for response in openai.ChatCompletion.create(
+                    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
                     ],
-                    stream=True,
                     # messages=[{'role': 'system', 'content': 'you are ChatGPT'}, {
-                    #     'role': 'user', 'content': prompt}],
-                    # stream=True,
+                    #     'role': 'user', 'content': prompt}], ## 这是单轮会话。
+                    stream=True,
                 ):
-                    full_response += response.choices[0].delta.get(
-                        "content", "")
-                    message_placeholder.markdown(full_response + "▌")
-                    # print('session completed!')
-                message_placeholder.markdown(full_response)
-            st.session_state.messages.append(
-                {"role": "assistant", "content": full_response})
-
-
-## load the local_KB PDF file. 
-# async def localKB_mode():
+                    # 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):
-    ### clear all the prior conversation.
-    # st.session_state.conversation = None
-    # st.session_state.chat_history = None
-    # st.session_state.messages = []
-    # message_placeholder = st.empty()
-    clear_all() ## reset the conversation.
-
+    # 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"])
+    # # 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 := st.chat_input("Say something"):
-    # prompt = st.chat_input("Say something")
-    # print('prompt now:', prompt)
-    # print('----------'*5)
     # if prompt:
-    if prompt := st.chat_input("Say something"):
+    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 = ""
 
-            # if radio_2 == "知识库模式":
-                # ! 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)
-                # print('QA_PROMPT:', QA_PROMPT)
-
+                ### 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)
@@ -317,47 +469,129 @@ def localKB_mode(username):
                 # # # index = load_index_from_storage(storage_context)
                 # print("storage_context:", storage_context)
                 # index = load_index_from_storage(storage_context,service_context=service_context)
-                storage_context = StorageContext.from_defaults(persist_dir=f"./{username}/")
-                print('storage_context:',storage_context)
-                index = load_index_from_storage(storage_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)
+                
+                ## 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(streaming=False, 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(llama_index_reply)
-                # 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})
+                # 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��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():
-    print('数据分析模式启动！')
+# 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}_upload.csv'
-    # uploaded_file_path = "/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/test_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)
     
-    # st.write(f"passed file path in data_mode: {uploaded_file_path}")
-    # tmp1 = pd.read_csv(uploaded_file_path)
-    # st.markdown('成功启动数据模式，以下是加载的文件内容')
-    # st.write(tmp1[:5])
-
     # Initialize chat history
     if "messages" not in st.session_state:
         st.session_state.messages = []
@@ -372,76 +606,130 @@ async def data_mode():
     prompt = st.chat_input("Say something")
     print('prompt now:', prompt)
     print('----------'*5)
-    # if prompt:
     if prompt:
-        st.session_state.messages.append({"role": "user", "content": prompt})
-        with st.chat_message("user"):
-            st.markdown(prompt)
-
-        with st.chat_message("assistant"):
-            async with CodeInterpreterSession() as session:
-                # 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. 你需要用提问的语言来回答（如：中文提问你就用中文来回答，英文提问你就用英文来回答）。
-                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)
-                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('Thinking...', 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, detailed_error=True)
-
-                    # 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。
-        # st.session_state.messages.append({"role": "assistant", "content": full_response})
-
+        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
@@ -456,8 +744,6 @@ authenticator = stauth.Authenticate(
     config['preauthorized']
 )
 
-
-###'''authentication with a remove cloud-based database.'''
 # authentication with a remove cloud-based database.
 # 导入云端用户数据库。
 
@@ -468,8 +754,10 @@ authenticator = stauth.Authenticate(
 
 # deta = Deta(DETA_KEY)
 
-# # mybase is the name of the database in Deta. You can change it to any name you want.
+# 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 = []
@@ -501,9 +789,13 @@ authenticator = stauth.Authenticate(
 #     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):
@@ -514,13 +806,6 @@ authenticator = stauth.Authenticate(
 #         st.success('注册成功！')
 # except Exception as e:
 #     st.error(e)
-''''''
-
-# user, authentication_status, username = authenticator.login('用户登录', 'main')
-# user, authentication_status, username = authenticator.login('用户登录', 'sidebar')
-user, authentication_status, username = authenticator.login('sidebar')
-
-# print("name", name, "username", username)
 
 if authentication_status:
     with st.sidebar:
@@ -528,187 +813,224 @@ if authentication_status:
             """
             <style>
             [data-testid="stSidebar"][aria-expanded="true"]{
-                min-width: 600px;
-                max-width: 600px;
+                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！_')
+        st.write(f'_Welcome and Hope U Enjoy Staying Here_')
         authenticator.logout('登出', 'sidebar')
 
-        ## 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)
+        # 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",
+        )
 
-        # 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():
-            st.markdown("#### 快速上手指南")
+        ## 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、且不支持长文。建议适当使用。")
+                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("* “联网模式”和“知识库模式”均基于检索功能，仅限一轮对话，不会保持之前的会话记录。")
+                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1-5分钟左右，期间需要保持网络畅通。")
+                    "* “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。")
-
-        st.markdown("#### 参考资料")
-        with st.expander(label="**核心模式的专用提示词Prompt示例**", expanded=False):
-            # with st.subheader(body="提示词Prompt"):
-            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2023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')
-
-    col1, col2 = st.columns(spec=[1, 2])
-    radio_2 = col2.radio(label='模式选择', options=[
-        '核心模式', '联网模式', '知识库模式', '数据模式'], horizontal=True, label_visibility='visible')
-    radio_1 = col1.radio(label='ChatGPT版本', options=[
-        'GPT-3.5', 'GPT-4.0'], horizontal=True, label_visibility='visible')
-
-elif authentication_status == False:
-    st.error('⛔ 用户名或密码错误！')
-elif authentication_status == None:
-    st.warning('⬅ 请先登录！')
-
-### 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:
-                # pdf_file = PyPDF2.PdfReader(uploaded_file)
-                PyPDF2.PdfReader(uploaded_file)
-                # st.write(pdf_file.pages[0].extract_text())
-                # 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 st.spinner('正在为您解析新知识库...请耐心等待'):
-                with spinner:
-                    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 🔃')
-                    # spinner = st.empty()
+                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。")
 
-            else:
-                if '.csv' in filename:
-                    csv_file = pd.read_csv(uploaded_file)
-                    csv_file.to_csv(f'./{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])
+        ## 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)
 
-                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())
-                    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.')
-        except Exception as e:
-            st.write(e)
+        ## 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)
 
-        ## 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.')
+        # 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')
 
-        return None
 
+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:
-        if radio_2 == "核心模式":
-            print(f'radio 选择了 {radio_2}')
-            # * 也可以用命令执行这个python文件。’streamlit run frontend/app.py‘
-            asyncio.run(text_mode())
-
-        if radio_2 == "联网模式":
-            print(f'radio 选择了 {radio_2}')
-            asyncio.run(text_mode())
-
-        if radio_2 == "知识库模式":
-            print(f'radio 选择了 {radio_2}')
-
-            path = f'./{username}/vector_store.json'
-            if os.path.exists(path):
-                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)
-                # st.markdown("注意：系统中已经存在一个知识库，您现在可以开始提问！")
-
-            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)
-
-            localKB_mode(username)
-            # asyncio.run(localKB_mode())
-
-        if radio_2 == "数据模式":
-            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)
-                asyncio.run(data_mode())
-    except:
-        # st.markdown('**请先登录！**')
-        pass
+        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

chatgpt.py

@@ -0,0 +1,61 @@
+
+# -*- coding: utf-8 -*-
+import requests
+import streamlit as st
+import openai
+from openai import OpenAI
+import os
+from dotenv import load_dotenv
+import numpy as np
+import pandas as pd
+import csv
+import tempfile
+from tempfile import NamedTemporaryFile
+import pathlib
+from pathlib import Path
+import re
+from re import sub
+import matplotlib.pyplot as plt
+from itertools import product
+from tqdm import tqdm_notebook, tqdm, trange
+import time
+from time import sleep
+import pretty_errors
+import seaborn as sns
+from matplotlib.pyplot import style
+from rich import print
+import warnings
+warnings.filterwarnings('ignore')
+
+load_dotenv()
+### 设置openai的API key
+os.environ["OPENAI_API_KEY"] = os.environ['user_token']
+openai.api_key = os.environ['user_token']
+
+##NOTE: 需要注意OpenAI的版本
+def chatgpt(user_prompt, sys_prompt="You are professional consultant", openai_model="gpt-3.5-turbo-16k"):
+    # print('chatgpt mode starts!')
+    try:
+        openai_client = OpenAI()
+        chatgpt_response = openai_client.chat.completions.create(
+            model=openai_model,
+            messages=[
+                {"role": "system", "content": sys_prompt},
+                {"role": "user", "content": user_prompt},
+            ],
+            stream=False,
+        )
+        
+        full_response = chatgpt_response.choices[0].message.content
+        print(full_response)
+    except Exception as e:
+        print(e)
+        full_response = "Sorry, I don't have an answer for that yet."
+    
+    # return chatgpt_response, full_response
+    return full_response
+
+# user_prompt = "where is Shanghai?"
+# system_prompt = "You are a helpful AI assistant."
+# full_response = chatgpt(user_prompt=user_prompt, sys_prompt=system_prompt)
+# print(full_response)

joeshi/database_name.csv

@@ -1,6 +1,24 @@
-name,date
-joeshi,2023年9月
-joeshi,2023年9月
-Web App开发需求说明 20230512.pdf,2023-09-24 20:59
-Sparks_of_AGI.pdf,2023-09-24 21:07
-AI与ChatGPT讲座大纲 20230615.pdf,2023-09-24 21:11
+大模型LLM解决方案调研问卷.pdf,2024-01-18 20:12
+joeshi_upload.pdf,2024-01-18 22:32
+joeshi_upload.pdf,2024-01-18 22:32
+joeshi_upload.pdf,2024-01-18 22:32
+专属ChatGPT产品介绍 20230415.pdf,2024-03-10 21:37
+专属ChatGPT产品介绍 20230415.pdf,2024-03-10 21:37
+Web App开发需求说明 20230512.pdf,2024-03-10 23:02
+Web App开发需求说明 20230512.pdf,2024-03-10 23:02
+Web App开发需求说明 20230512.pdf,2024-03-10 23:02
+大模型LLM解决方案调研问卷.pdf,2024-03-12 12:51
+大模型LLM解决方案调研问卷.pdf,2024-03-12 12:51
+企业数据资源相关会计处理暂行规定.pdf,2024-03-12 12:51
+企业数据资源相关会计处理暂行规定.pdf,2024-03-12 12:54
+企业数据资源相关会计处理暂行规定.pdf,2024-03-12 12:54
+企业数据资源相关会计处理暂行规定.pdf,2024-03-12 12:54
+joeshi_upload.pdf,2024-03-12 18:19
+joeshi_upload.pdf,2024-03-12 18:19
+joeshi_upload.pdf,2024-03-12 18:20
+企业数据资源相关会计处理暂行规定.pdf,2024-03-12 21:53
+企业数据资源相关会计处理暂行规定.pdf,2024-03-12 22:40
+企业数据资源相关会计处理暂行规定.pdf,2024-03-12 22:54
+企业数据资源相关会计处理暂行规定.pdf,2024-03-12 22:57
+AI与ChatGPT讲座大纲 20230615.pdf,2024-03-13 13:20
+AI与ChatGPT讲座大纲 20230615.pdf,2024-03-13 13:20

joeshi/default__vector_store.json

@@ -0,0 +1 @@
+{"embedding_dict": {"2d707a34-28f5-4cfa-8973-961c2f7e96e1": [-0.010600954294204712, -0.007286855950951576, 0.010656420141458511, -0.02024511806666851, 0.015849431976675987, 0.002634638687595725, -0.003260365454480052, 0.02477947063744068, -0.014032918028533459, -0.03535962477326393, 0.003071434097364545, 0.027802372351288795, -0.021604038774967194, -0.0021302441600710154, -0.0038583590649068356, -0.003899958683177829, -0.010677220299839973, -0.002721304539591074, 0.019163530319929123, -0.0012913196114823222, -0.01995392143726349, 0.024696271866559982, -0.01525317132472992, 0.0017263817135244608, -0.0022775758989155293, 0.010781219229102135, 0.025389596819877625, -0.015669167041778564, 0.001039988943375647, 0.012757197953760624, -0.0007518253987655044, -0.018484070897102356, -0.02182590216398239, -0.0011015216587111354, -0.006662862841039896, -0.016071297228336334, 0.00825751293450594, -0.00332276476547122, 0.02959115244448185, -0.010115626268088818, 0.009394567459821701, 0.01849793829023838, 0.024765605106949806, 0.0011665209894999862, -0.02265789359807968, 0.010143359191715717, 0.008992438204586506, -0.018026474863290787, 0.008493243716657162, 0.01518383901566267, -0.015336371026933193, 0.021340573206543922, -0.009020171128213406, -0.0029050358571112156, -0.006787661463022232, 0.006770328152924776, -0.0064444649033248425, 0.020078720524907112, -0.011266547255218029, 0.0006651596049778163, 0.012251070700585842, 0.0006075269193388522, -0.029230624437332153, -0.007321522571146488, -0.021437639370560646, 0.007910849526524544, -0.02695651538670063, 0.00843084417283535, 0.006077002268284559, 0.017097419127821922, 0.018428605049848557, 0.013055328279733658, -0.0025150401052087545, -0.008950838819146156, 0.041821423918008804, -0.006666329223662615, -0.014573711901903152, -0.013804120011627674, -0.008084180764853954, 0.010108692571520805, 0.021604038774967194, -0.020064854994416237, 0.007536453660577536, 0.01512837316840887, 0.007460187654942274, 0.029286090284585953, 0.004638351034373045, -0.006378598976880312, -0.014490513131022453, 0.0010469222906976938, 0.00860417541116476, 0.0031806330662220716, 0.020494716241955757, 0.010587087832391262, -0.017125152051448822, 0.03036767803132534, -0.002047044923529029, -0.0005160945584066212, -0.00717592379078269, -0.03996330872178078, 0.007702851668000221, -0.019537925720214844, -0.031698863953351974, -0.0023642415180802345, -0.041655026376247406, 0.012479867786169052, 0.03474950045347214, 0.021007778123021126, 0.01412305049598217, 0.016334760934114456, -0.01514223963022232, 0.028897827491164207, -0.008819106966257095, -0.017069686204195023, -0.005886337719857693, -0.009193502366542816, 0.015932630747556686, 0.0007947249105200171, 0.0014265181962400675, -0.008146580308675766, 0.020439250394701958, -0.01324945967644453, -0.014878775924444199, -0.01346439030021429, 0.004714616574347019, -0.003893025452271104, -0.014781709760427475, -0.02577785961329937, 0.033335112035274506, -0.01240360178053379, -0.008091114461421967, 0.00029748017550446093, 0.010594021528959274, -0.0027507708873599768, -0.04068436846137047, 0.020037120208144188, -0.025958124548196793, 0.0033695644233375788, -0.03635801374912262, -0.016071297228336334, 0.003315831534564495, 0.035914286971092224, -0.011446812190115452, -0.01568303443491459, 0.00614286819472909, 0.025070667266845703, -0.008312978781759739, 0.021035511046648026, 0.008846839889883995, 0.017527280375361443, 0.01334652490913868, -0.02081364579498768, 0.00965109746903181, 0.040018774569034576, 0.012001472525298595, -0.009034037590026855, 0.012667065486311913, 0.02945248782634735, -0.005345543380826712, -0.009734297171235085, -0.011828141286969185, 0.03025674633681774, 0.01145374495536089, 0.013083061203360558, 0.018387004733085632, 0.02672078274190426, -0.0029206357430666685, 0.011495345272123814, 0.01521157193928957, 0.007966315373778343, -0.014559845440089703, 0.003267298685386777, -0.0209939107298851, 0.0007713251397944987, 0.012687865644693375, -0.009089503437280655, 0.013041461817920208, -0.0012765864375978708, -0.025112267583608627, -0.020203519612550735, -0.015059040859341621, -0.006548463832587004, 0.012008406221866608, 0.008819106966257095, -0.01989845559000969, -0.0064375316724181175, 0.0009654564200900495, -0.014698510989546776, 0.01419238280504942, -0.03480496630072594, 0.035692423582077026, 0.019024865701794624, 0.010649487376213074, -0.019621126353740692, -0.5959275364875793, -0.018858466297388077, 0.011925207450985909, -0.014767843298614025, -0.0073839216493070126, 0.0068535273894667625, 0.0012531867250800133, 0.005446075461804867, -0.006420198827981949, -0.006382065825164318, -0.029008759185671806, 0.0011899207020178437, 0.003275965340435505, -0.007896983064711094, -0.011717209592461586, -0.01602969691157341, -0.006742595229297876, 0.003518629353493452, 0.006628196686506271, 0.03028447926044464, -0.041682757437229156, 0.0055188750848174095, 0.01433798111975193, 0.0045447517186403275, -0.01118334848433733, 0.01406758464872837, 0.022463761270046234, 0.0028097035828977823, -0.00432982062920928, 0.004437286406755447, -0.046868838369846344, 0.009817495942115784, 0.004981547128409147, -0.013041461817920208, 0.04575951397418976, -0.015308638103306293, -0.018913932144641876, 0.008839906193315983, 0.009644164703786373, 0.018997132778167725, -0.012895863503217697, -0.01806807518005371, 0.004253555089235306, 0.010802019387483597, -0.0040143574588000774, 0.0031858328729867935, 0.030811406672000885, -0.005983403418213129, -0.02348988503217697, 0.004569018259644508, 0.0036330281291157007, -0.01761048100888729, 0.00837537832558155, -0.00702685909345746, 0.006642063148319721, 0.0010451889829710126, 0.034971363842487335, -0.047673095017671585, 0.0024803737178444862, -0.016501158475875854, -0.001149187795817852, 0.010850551538169384, 0.0042986213229596615, -0.02437734231352806, -0.022491494193673134, 0.019565658643841743, -0.015585968270897865, -0.008118847385048866, 0.018650468438863754, -0.004381820093840361, -0.014268648810684681, 0.012999862432479858, 1.0467597348906565e-05, -0.019884590059518814, -0.009006304666399956, 0.03613615036010742, 0.006624729838222265, 0.0028807695489376783, -7.458670734195039e-05, 0.025334130972623825, 0.004506618715822697, 0.0009663230739533901, 0.027233844622969627, -0.00037937931483611465, 0.004517018795013428, 0.005952203646302223, -0.012050005607306957, -0.011869740672409534, 0.016875553876161575, -0.0056506069377064705, -0.0018199806800112128, 0.02864822931587696, -0.007453254424035549, -0.009886828251183033, -0.0015660501085221767, 0.012701732106506824, -0.01608516275882721, -0.025736261159181595, -0.000288380280835554, -0.02196456678211689, 0.007945516146719456, -0.02939702197909355, 0.015267038717865944, 0.027594374492764473, 0.037633735686540604, 0.027053579688072205, -0.03294685110449791, -0.020439250394701958, 0.01820673979818821, -0.008479377254843712, -0.018081942573189735, -0.005456475540995598, 0.0021787770092487335, -0.029702086001634598, 0.006146335043013096, -0.027705306187272072, 0.01327719260007143, -0.004596751183271408, 0.003206632798537612, -0.0034666298888623714, 0.006780728232115507, -0.015988096594810486, 0.01632089354097843, -0.00620873412117362, 0.02150697261095047, 0.03228125721216202, 0.01205693930387497, -0.018969399854540825, 0.025486662983894348, 0.02416934445500374, -0.005962603725492954, -0.035692423582077026, 0.03857665881514549, -0.013845720328390598, 0.047673095017671585, 0.007564186584204435, 0.024724004790186882, 0.0021978432778269053, 0.004558618180453777, -0.009158836677670479, -0.025555996224284172, -0.004881015047430992, 0.02615225687623024, 0.003033301094546914, -0.011301213875412941, -0.02325415425002575, -0.01605742983520031, -0.005574340932071209, 0.0017142485594376922, -0.017596613615751266, -0.003470096504315734, -0.017943276092410088, 0.022436028346419334, 0.04073983430862427, -0.003920758608728647, 0.0018529137596488, 0.00931830145418644, -0.03533189371228218, -0.0013337858254089952, -0.008153514005243778, -0.006714862305670977, 0.01973205804824829, -0.003667694516479969, -0.015544368885457516, -0.006239933893084526, -0.0191496629267931, -0.009567898698151112, 0.02518160082399845, -0.030950071290135384, -0.016015829518437386, 0.01218867115676403, 0.00851404294371605, -0.019607258960604668, 0.0021441106218844652, -0.005844737868756056, -0.021298974752426147, 0.0022082433570176363, -0.017582746222615242, 0.021631771698594093, -0.010094826109707355, 0.00037092939601279795, 0.027746906504034996, -0.009796695783734322, -0.0008454243652522564, 0.003750893520191312, -0.0026675716508179903, -0.0015851164935156703, 0.022172564640641212, -0.017457948997616768, 0.024502141401171684, -0.01114174909889698, 0.031587932258844376, -0.03203165903687477, 0.03375110775232315, -0.008874572813510895, 0.046009112149477005, 0.009380700998008251, -0.02500133588910103, -0.017083551734685898, 0.006434065289795399, -0.002868636278435588, 0.014920375309884548, -0.003274231916293502, -0.007113524712622166, 0.013755587860941887, -0.014393447898328304, 0.018137408420443535, -0.011301213875412941, 0.014448913745582104, -0.012334269471466541, 0.011918273754417896, 0.010947617702186108, -0.024585340172052383, 0.005317810457199812, -0.01409531757235527, 0.010815885849297047, -0.007841517217457294, 0.007397788111120462, -0.022463761270046234, -0.012542267329990864, -0.01139827910810709, -0.003778626676648855, 0.01220253761857748, 0.017901677638292313, -0.013956652022898197, 0.02733091078698635, -0.013180127367377281, 0.017014220356941223, 0.01731928251683712, -0.014213182963430882, -0.004569018259644508, -0.0006361266132444143, 0.011023882776498795, 0.01225800346583128, -0.003653828054666519, -0.015530502423644066, 0.03039541095495224, -0.015322504565119743, 0.03278045356273651, -0.0041564893908798695, -0.0094223003834486, 0.006680195685476065, 0.02228349819779396, -0.04961440712213516, 0.011613210663199425, 0.0021926434710621834, 0.010233491659164429, 0.010261224582791328, -0.01952406018972397, 0.02862049639225006, -0.03294685110449791, 0.02961888536810875, -0.024266408756375313, -0.042043287307024, 0.04831095412373543, -0.007141257636249065, 0.0024422407150268555, 0.00602153642103076, -0.007987115532159805, 0.017915543168783188, -0.002539306413382292, 0.019856857135891914, 0.005390609614551067, 0.022006167098879814, 0.011273480951786041, -0.023822680115699768, -0.0026138389948755503, -0.011481478810310364, -0.010372157208621502, -0.0169171541929245, 0.03286365047097206, -0.018941666930913925, 0.017873944714665413, -0.0095054991543293, 0.014365714974701405, 2.046124245680403e-05, 0.006250333972275257, 0.027247712016105652, -0.006503397598862648, 0.023351218551397324, -0.009151902981102467, -0.023101622238755226, -0.0010529888095334172, 0.018220607191324234, -0.019995521754026413, -0.012972129508852959, -0.009443099610507488, -0.018276073038578033, -0.03111647069454193, -0.005078612826764584, -0.01114174909889698, -0.013214793056249619, -0.010344423353672028, 0.015932630747556686, -0.020175786688923836, 0.011432945728302002, -0.013138527050614357, -0.00558474101126194, -0.032530855387449265, -0.008396177552640438, 0.009040971286594868, -0.008389244787395, -0.044289663434028625, -0.02075817994773388, 0.001915313070639968, -0.0169726200401783, -0.04903201386332512, -0.012119337916374207, -0.0031650331802666187, -0.007370055187493563, 0.026485051959753036, -0.01645955815911293, -0.00620873412117362, -0.03710680827498436, 0.011939073912799358, -0.0020297118462622166, -0.010684153065085411, -0.014837176539003849, 0.03655214607715607, 0.005446075461804867, -0.03022901341319084, -0.01849793829023838, -0.02064724825322628, 0.021687237545847893, 0.04323580861091614, 0.05502235144376755, 0.007896983064711094, -0.0013441857881844044, -0.016251560300588608, 0.01573850028216839, -0.02939702197909355, -0.04526032134890556, 0.01119721494615078, -0.0021770435851067305, -0.014948108233511448, -0.009034037590026855, 0.006683662533760071, 0.005851671565324068, 0.013928919099271297, 0.003173699602484703, 0.021285107359290123, -0.015447302721440792, -0.008139647543430328, 0.008014848455786705, -0.012708665803074837, -0.0020435783080756664, 0.02942475490272045, 0.006125535350292921, -0.0073908548802137375, -0.009956161491572857, 0.0231154877692461, 0.017901677638292313, -0.0033816974610090256, -0.011044682934880257, -0.017984876409173012, -0.0015140506438910961, 0.005962603725492954, -0.010233491659164429, -0.010344423353672028, 0.04406780004501343, -0.029757551848888397, -0.001527917105704546, -0.0013866520021110773, 0.015988096594810486, 0.012729465030133724, 0.03289138525724411, 0.011259614489972591, -0.030866872519254684, 0.01118334848433733, -0.014767843298614025, 0.015558235347270966, 0.0075503201223909855, -0.011966806836426258, -0.0047042169608175755, 0.012604666873812675, 0.025985857471823692, -0.02437734231352806, 0.00665939599275589, -0.0004952947492711246, 0.0036226282827556133, -0.0033435646910220385, -0.015849431976675987, 0.013623856008052826, -0.040767569094896317, -0.01774914562702179, -0.033251915127038956, -0.012105471454560757, -0.01726381666958332, 0.008347644470632076, 0.004988480359315872, -0.0211048424243927, -0.03228125721216202, -0.0037959597539156675, 0.015308638103306293, 0.00843084417283535, -0.017971009016036987, -0.029147423803806305, -0.012348135933279991, 0.00567833986133337, -0.007023392245173454, 0.02376721426844597, 0.017901677638292313, -0.017014220356941223, 0.008299112319946289, 0.01995392143726349, -0.037467338144779205, 0.007037258706986904, -0.02494587004184723, -0.003071434097364545, 0.007037258706986904, -0.011127881705760956, -0.009172703139483929, 0.004499685484915972, 0.014448913745582104, -0.002736904425546527, -0.002509840065613389, -0.0013138527283445, 0.013707054778933525, 0.011114015243947506, -0.001543516991659999, -0.008923105895519257, -0.004073290154337883, 0.02371174842119217, -0.007196723949164152, -0.02609679102897644, -0.04797815904021263, 0.004541285336017609, -0.0337233766913414, 0.016473425552248955, 0.027774639427661896, 0.017527280375361443, 0.0030125014018267393, 0.008423910476267338, -0.004589817952364683, 0.02956341952085495, 0.0034302303101867437, 0.0018719801446422935, -0.01946859434247017, 0.011100148782134056, 0.009893761947751045, 0.017513414844870567, 0.026623718440532684, 0.010982283391058445, -0.02059178240597248, 0.03522096201777458, -0.022352829575538635, 0.0209939107298851, 0.04584271460771561, -0.015364103950560093, 0.039214517921209335, -0.004721549805253744, -0.012965195812284946, -0.03524869307875633, -0.007480987347662449, -0.0047388831153512, -0.004523952025920153, 0.004523952025920153, -0.013637722469866276, -0.01605742983520031, -0.01392198633402586, -0.01657049171626568, 0.0075503201223909855, 0.0013511190190911293, -0.01975979097187519, -0.023989079520106316, -0.010836685076355934, -0.005505008157342672, 0.012985995970666409, 0.034999094903469086, -0.03125513717532158, -0.017305416986346245, -0.011495345272123814, 0.023864280432462692, 0.042653415352106094, -0.012916662730276585, 0.014545978978276253, -0.021756568923592567, -0.01869206875562668, -0.00714819086715579, -0.017430216073989868, -0.0007830250542610884, 0.004080223385244608, 0.01562756858766079, -0.01608516275882721, 0.037439603358507156, -9.690313891042024e-05, -0.002924102358520031, 0.025417331606149673, 0.001367585500702262, 0.001363252173177898, -0.0054842084646224976, -0.013013728894293308, -0.0379110649228096, 0.004319421015679836, 0.006492997985333204, 0.0105662876740098, 0.023697882890701294, -0.013187060132622719, -0.023184821009635925, 0.005581274162977934, -0.013207860291004181, 0.02285202406346798, -0.013422790914773941, -0.0546618215739727, -0.03189299628138542, -0.004881015047430992, -0.03192072734236717, -0.0045378184877336025, -0.03552602231502533, -0.01616836152970791, 0.024502141401171684, -0.003823692910373211, 0.02563919499516487, 0.000946389976888895, 0.01597423106431961, -0.03616388514637947, -0.016501158475875854, 0.006250333972275257, 0.013443591073155403, 0.002707438077777624, -0.00943616684526205, -0.039380915462970734, 0.0026953048072755337, 0.01602969691157341, 0.02053631655871868, 0.04467792809009552, -0.0032846317626535892, -0.014795576222240925, 0.01573850028216839, 0.016209961846470833, -0.019316062331199646, -0.01146761141717434, -0.0009593898430466652, -0.02583332546055317, -0.01648729294538498, -0.018400872126221657, -0.0068431273102760315, -0.005591674242168665, -0.007786050904542208, 0.02145150676369667, -0.013762520626187325, 0.026027457788586617, -0.011848941445350647, 0.002657171804457903, 0.011980673298239708, 0.005300477147102356, 0.055327415466308594, -0.012687865644693375, 0.027691440656781197, 0.048505086451768875, 0.002426641061902046, -0.015516635961830616, -0.023032288998365402, 0.011148681864142418, 0.000750525388866663, 0.022033900022506714, 0.02792717143893242, -0.0001294569665333256, -0.014698510989546776, 0.029313823208212852, 0.026540519669651985, -0.02405841276049614, -0.016612090170383453, -0.014171583577990532, 0.008950838819146156, -0.005882870871573687, -0.028842361643910408, -0.02494587004184723, 0.007453254424035549, 0.0026086389552801847, -0.0035671622026711702, 0.010649487376213074, -0.014421180821955204, -0.02167337015271187, -0.015516635961830616, 0.011800408363342285, 0.014213182963430882, 0.011855874210596085, 0.008423910476267338, 0.002665838459506631, 0.022408295422792435, -0.002790637081488967, -0.013152394443750381, 0.022172564640641212, -0.0035238293930888176, 0.026166122406721115, -0.01432411465793848, 0.003716227365657687, 0.006829260848462582, 0.015835566446185112, -0.0006595263257622719, -0.004579418338835239, 0.009866029024124146, 0.045066189020872116, -0.01731928251683712, -0.003969291225075722, 0.00042444549035280943, -0.00759885273873806, 0.0505295991897583, -0.015017440542578697, 0.003934625070542097, 0.0003408130432944745, -0.032447654753923416, -0.009727363474667072, 0.024682404473423958, 0.010240424424409866, -0.0005056946538388729, 0.009034037590026855, -0.008229779079556465, -0.01946859434247017, 0.0002040978433797136, 0.010171092115342617, -0.004693816881626844, -0.01786007732152939, 0.009866029024124146, 0.00517567852512002, -0.008666574954986572, 0.005459942389279604, -0.00661779660731554, -0.011530010960996151, -0.004378353711217642, 0.02835703268647194, 0.0014282515039667487, 0.0012332536280155182, -0.0004476285830605775, -0.02752504125237465, -0.029951682314276695, 0.011578544043004513, -0.0035220959689468145, -0.01033749058842659, 0.026845581829547882, -0.031615663319826126, -0.02244989573955536, -0.02416934445500374, 0.009789763018488884, 0.02735864371061325, -0.02185363508760929, 0.026221588253974915, 0.018331538885831833, 0.008132713846862316, 0.0006157601601444185, -0.012888929806649685, 0.00432982062920928, 0.011010016314685345, -0.005498074926435947, -0.0019031798001378775, -0.0013173193437978625, -0.014545978978276253, 0.03477723151445389, -0.01406758464872837, 0.00946389976888895, -0.00611513527110219, -0.04318034276366234, -0.02362854965031147, 0.006707929074764252, 0.0044303531758487225, -0.020522449165582657, -0.008000981993973255, -0.025958124548196793, -0.009387633763253689, -0.010191892273724079, -0.001462917891331017, 0.001115388236939907, -0.0011795208556577563, 0.017152884975075722, 0.01519770547747612, 0.0026311720721423626, 0.008978571742773056, -0.008971638046205044, -0.011134815402328968, -0.02408614568412304, -0.010600954294204712, -0.047562163323163986, -0.021576303988695145, 0.014656911604106426, 0.03369564190506935, 0.022172564640641212, -0.030700474977493286, -0.012625466100871563, -0.018400872126221657, -0.05563247576355934, -0.031532466411590576, -0.010850551538169384, 0.03136606886982918, -0.013408924452960491, 0.01651502586901188, 0.004062890075147152, 0.01648729294538498, -0.0007557253120467067, 0.02755277417600155, 0.007321522571146488, -0.0041391560807824135, 0.009269768372178078, -0.025680795311927795, 0.01995392143726349, -0.006905526854097843, -0.046868838369846344, -0.010892150923609734, 0.006336999591439962, 0.013159327208995819, 0.0015383170684799552, 0.025029068812727928, -0.009068704210221767, -0.01224413700401783, 0.0012358535313978791, 0.00038847920950502157, -0.009762030094861984, 0.005810071714222431, 0.017055818811058998, -0.05241544544696808, 0.0012349869357421994, 0.012362002395093441, -0.002346908440813422, -0.014767843298614025, -0.003983157686889172, -0.009221235290169716, -0.019745923578739166, -0.0032274324912577868, -0.01143987849354744, -0.005057813134044409, 0.011010016314685345, 0.012618533335626125, 0.04467792809009552, 0.004714616574347019, -0.008354578167200089, 0.045149389654397964, 0.022006167098879814, -0.010399890132248402, -0.001309519517235458, 0.01141214556992054, -0.024308009073138237, -0.012167870998382568, 0.02709518000483513, -0.013741721399128437, 0.014448913745582104, -0.007564186584204435, 0.011495345272123814, -0.015988096594810486, -0.023503750562667847, -0.006375132594257593, -0.006132468581199646, -0.02287975698709488, 0.012639332562685013, 0.034167103469371796, -0.02159017138183117, 0.011495345272123814, -0.01616836152970791, -0.01145374495536089, 0.016043564304709435, -0.00042812881292775273, -0.019454726949334145, -0.01602969691157341, -0.0022862425539642572, 0.007224456872791052, 0.009179635904729366, 0.003792493138462305, 0.014961974695324898, 0.01428251527249813, -0.010288957506418228, 0.015766233205795288, 0.25891566276550293, -0.014157717116177082, 0.017041953280568123, -0.002119844313710928, 0.014837176539003849, 0.013741721399128437, 0.03136606886982918, 0.025445064529776573, -0.0006772928172722459, -0.0055258083157241344, -0.009429233148694038, 0.0026675716508179903, 0.015613701194524765, -0.008292178623378277, 0.006097801961004734, -0.036663077771663666, -0.024613073095679283, -0.04035157337784767, -0.005882870871573687, -0.00848631002008915, -0.01935766078531742, -0.014213182963430882, 0.011432945728302002, -0.007245256565511227, -0.006045802496373653, 0.003447563387453556, -0.014684644527733326, -0.016251560300588608, 0.030672742053866386, 0.0033279648050665855, -0.020522449165582657, 0.004305554553866386, 0.0035012962762266397, 0.0168894212692976, -0.008770573884248734, -0.013873453252017498, 0.024044545367360115, 0.0032655654940754175, 0.022602427750825882, -4.067873669555411e-05, 0.007876182906329632, -0.010635620914399624, 0.013720921240746975, -0.050890129059553146, 0.02330962009727955, 0.015447302721440792, -0.027247712016105652, -0.013665455393493176, -0.00571993924677372, 0.011758808977901936, -0.009075636975467205, 0.013609989546239376, 0.025486662983894348, 0.033279646188020706, -0.012077738530933857, 0.00943616684526205, -0.016986487433314323, 0.0076820519752800465, -0.018414737656712532, 0.006357799284160137, -0.052775975316762924, 0.023961346596479416, 0.0029275689739733934, 0.010628687217831612, -0.02167337015271187, 0.0022533093579113483, -0.016099030151963234, -0.023184821009635925, 0.016043564304709435, -0.023503750562667847, -0.0024214410223066807, -0.01568303443491459, -0.005834338255226612, 0.006409798748791218, -0.0506405308842659, -0.05166665092110634, 0.026679184287786484, 0.013505990616977215, 0.013429724611341953, 0.057074595242738724, -0.0022533093579113483, -0.011114015243947506, 0.01408145111054182, -0.01614062860608101, -0.0031303667929023504, -0.03455536812543869, 0.03480496630072594, -0.0013424524804577231, -0.012382802553474903, -0.009588698856532574, -0.02265789359807968, 0.003972758073359728, 0.0014005184639245272, -0.015544368885457516, 0.0067391288466751575, 0.02067498117685318, 0.01989845559000969, -0.004101023077964783, -0.015267038717865944, 0.012341202236711979, -0.026082923635840416, 0.06162281334400177, 0.04046250507235527, 0.0169171541929245, -0.014421180821955204, -0.008909239433705807, -0.008548709563910961, 0.01231346931308508, 0.014643045142292976, -0.009720430709421635, -0.007467120885848999, -0.013395057991147041, 0.010122559033334255, 0.0009853895753622055, 0.014012117870151997, 0.01941312849521637, 0.0148233100771904, 0.0025462396442890167, 0.008638842031359673, -0.020175786688923836, 0.027053579688072205, -0.034111637622117996, -0.002189176855608821, 0.01602969691157341, -0.016015829518437386, -0.030811406672000885, -0.009512432850897312, -0.01815127395093441, -0.01975979097187519, -0.01515610609203577, 0.002034911885857582, -0.041738223284482956, 0.015280905179679394, -0.013096927665174007, -0.007113524712622166, 0.020744314417243004, 0.008146580308675766, 0.011599344201385975, 0.0042882212437689304, 0.008784440346062183, 0.0028738363180309534, 0.008722040802240372, -0.01521157193928957, 0.005272744223475456, 0.017582746222615242, -0.016625957563519478, 0.015267038717865944, 0.001215920434333384, 0.019801389425992966, -0.013110794126987457, -0.03272498771548271, -0.008306045085191727, -0.005116745829582214, 0.00513407913967967, 0.028481831774115562, 0.009165769442915916, -0.02443280816078186, -0.01608516275882721, 0.013658521696925163, -0.00430208770558238, -0.041849154978990555, -0.017596613615751266, 0.03369564190506935, -0.023503750562667847, -0.014961974695324898, -0.004919148050248623, -0.1776023805141449, 0.004534352105110884, 0.005948736798018217, -0.013214793056249619, 0.026776250451803207, 0.02695651538670063, 0.012951329350471497, 0.00824364647269249, 0.012063872069120407, 0.002651971997693181, 0.023753348737955093, -0.005560474470257759, 0.0012965196510776877, 0.0025913058780133724, 0.014490513131022453, 0.01229960285127163, -0.026942647993564606, 0.005158345215022564, 0.022838158532977104, 0.009907628409564495, 0.019912322983145714, -0.014545978978276253, -0.009602565318346024, 0.017707545310258865, 0.0001432151475455612, 0.00153225043322891, -0.020286718383431435, -0.004364487249404192, -0.007959382608532906, -0.027233844622969627, -0.007467120885848999, -0.009020171128213406, 0.009075636975467205, 0.015003574080765247, 0.03380657359957695, 0.003948491532355547, 0.004926081281155348, -0.025888793170452118, -0.02609679102897644, 0.008888439275324345, 0.02142377384006977, 0.026249321177601814, 0.011405212804675102, 0.010032427497208118, -0.011086282320320606, 0.023531483486294746, 0.011744942516088486, -0.02030058577656746, 0.010288957506418228, 0.0009567898814566433, 0.025417331606149673, -0.021784301847219467, 0.02239442989230156, 0.004825548734515905, 0.008507110178470612, 0.038327060639858246, 0.015988096594810486, -0.0013615188654512167, -0.002284509129822254, -0.004551684949547052, 0.009124170057475567, -0.0036711611319333315, -0.0074255215004086494, -0.012382802553474903, 0.005709539633244276, -0.020910711959004402, -0.0336124449968338, 0.013665455393493176, -0.012840397655963898, 0.007397788111120462, -0.00661779660731554, -0.027677573263645172, 0.017360882833600044, -0.011627077125012875, 0.028925560414791107, 0.020439250394701958, -0.02612452395260334, 0.01519770547747612, 0.03031221218407154, -0.005355943460017443, -0.009179635904729366, -0.0030159680172801018, -0.007973249070346355, 0.004336753860116005, -0.011710275895893574, -0.01643182523548603, 0.018858466297388077, -0.009255901910364628, 0.015488903038203716, -0.011280413717031479, 0.002937968820333481, -0.00761271920055151, -0.0032568988390266895, 0.00861804187297821, 0.024100011214613914, 0.01433798111975193, -0.018428605049848557, -0.0052276779897511005, -0.014116116799414158, -0.02056404948234558, 0.0056332736276090145, 0.006676729302853346, -0.012798797339200974, -0.007730584591627121, 0.027746906504034996, 0.012923596426844597, 0.015086773782968521, 0.035942018032073975, 0.050890129059553146, -0.005494608543813229, -0.043513137847185135, 0.003962357994168997, 0.02161790430545807, 0.029369289055466652, 0.0049780807457864285, 0.023143220692873, -0.030007148161530495, -0.0378555990755558, -0.012930530123412609, -0.005258877761662006, 0.019052598625421524, -0.010864417999982834, 0.021770436316728592, 0.03549829125404358, -0.012923596426844597, -0.01334652490913868, -0.05158345401287079, -0.015031307004392147, 0.01949632726609707, 0.0338897742331028, -0.010815885849297047, 0.03696814179420471, -0.005407942924648523, 0.02443280816078186, -0.0031650331802666187, 0.027941036969423294, -0.009782829321920872, -0.029896216467022896, -0.014795576222240925, -0.0009351234184578061, 0.01132894679903984, -0.002267176052555442, 0.007058058399707079, -0.03522096201777458, -0.015350237488746643, 0.026332521811127663, 0.002282775705680251, -0.018456337973475456, 0.016847820952534676, -0.03816066309809685, -0.009047904051840305, -0.014892642386257648, -0.034028440713882446, -0.001543516991659999, 0.01616836152970791, -0.0027871704660356045, 0.03555375710129738, -0.034860432147979736, 0.0094500333070755, -0.038354795426130295, 0.011509211733937263, -0.005047413054853678, -0.027663707733154297, -0.01737475022673607, 0.045066189020872116, -0.013901186175644398, -0.007120457943528891, -0.002086911117658019, 0.01406758464872837, -0.026055190712213516, -0.02373948134481907, 0.002242909511551261, -0.031033271923661232, 0.024266408756375313, -0.01016415935009718, 5.037446680944413e-05, -0.0035532957408577204, 0.011946006678044796, -0.021021643653512, -0.009845228865742683, 0.022297363728284836, -0.005231144372373819, 0.006687129382044077, 0.026734650135040283, -0.0074255215004086494, -0.026249321177601814, -0.0025913058780133724, -0.017901677638292313, -0.021950701251626015, 0.008028714917600155, 0.003574095433577895, 0.02830156683921814, -0.004756216425448656, -0.0167507566511631, 0.013665455393493176, -0.00012490700464695692, -0.008860706351697445, 0.03283591940999031, -0.006520730908960104, 0.013741721399128437, -0.037800133228302, -0.007702851668000221, -0.010455355979502201, -0.006981792859733105, 0.005726872477680445, -0.006614329759031534, 0.002431840868666768, -0.027941036969423294, 0.008888439275324345, 0.00190664641559124, -0.0035706288181245327, 0.016043564304709435, 0.0063231331296265125, -0.014518246054649353, 0.012133204378187656, -0.05546607822179794, 0.004000490996986628, 0.004517018795013428, -0.001111054909415543, -0.020466983318328857, -0.00483248196542263, 0.007300722878426313, -0.0006785928271710873, -0.004042090382426977, -0.010864417999982834, 0.02411387860774994, -0.01417851634323597, -0.036635346710681915, -0.10355517268180847, 0.02333735302090645, 0.0024370409082621336, 0.0056228735484182835, 0.007085791788995266, 0.007446321193128824, 0.0013641188852488995, 0.00016520658391527832, -0.005449542310088873, 0.001329452614299953, -0.005290077533572912, 0.005737272556871176, -0.015072907321155071, -0.012764131650328636, -0.014448913745582104, -0.015877164900302887, 0.011280413717031479, 0.016209961846470833, 0.016625957563519478, -0.006215667352080345, 0.016237694770097733, 0.0071343244053423405, 0.018733669072389603, 0.03685721009969711, 0.011107082478702068, 0.008278312161564827, -0.02862049639225006, 0.03211485967040062, -0.010288957506418228, -0.010600954294204712, -0.011668676510453224, -0.013408924452960491, 0.03616388514637947, -0.0012219870695844293, -0.00825751293450594, -0.01221640408039093, 0.012667065486311913, 0.013956652022898197, 0.02121577598154545, 0.03436123579740524, -0.010191892273724079, -0.021409906446933746, 0.013311859220266342, -0.016931021586060524, -0.02781623788177967, 0.0026311720721423626, -0.0059140706434845924, -0.018276073038578033, 0.0334460474550724, -0.03397297486662865, 0.02859276346862316, 0.010060160420835018, -0.014490513131022453, -0.030811406672000885, -0.002119844313710928, -0.007141257636249065, 0.026221588253974915, 0.013582255691289902, 0.00045152855454944074, 0.007633519358932972, 0.022560827434062958, 0.009124170057475567, 0.03222579136490822, -0.010843618772923946, -0.0014057183871045709, 0.011578544043004513, -0.027968769893050194, -0.025334130972623825, 0.009942295029759407, -0.035082295536994934, -0.008077247999608517, -0.0030853005591779947, 0.011100148782134056, 0.04470565915107727, -0.012126271612942219, -0.0032967650331556797, -0.03133833408355713, -0.0009602564969100058, -0.010372157208621502, 0.032503120601177216, 0.028925560414791107, 0.023503750562667847, -0.02081364579498768, 0.016501158475875854, 0.006427132058888674, 0.04037930443882942, -0.013866519555449486, 0.003385164076462388, -0.020827513188123703, -0.0026485053822398186, -0.012022272683680058, 0.003868758911266923, -0.004974613897502422, -0.003055834211409092, 0.0042916880920529366, 0.02394747920334339, -0.012660132721066475, 0.01780461147427559, 0.02233896404504776, 0.021021643653512, 0.025736261159181595, 0.01605742983520031, 0.006714862305670977, -0.00861804187297821, -0.032558590173721313, -0.008909239433705807, -0.04132222756743431, -0.02319868840277195, 0.009526299312710762, -0.006638596300035715, 0.012687865644693375, -0.002166643738746643, -0.0036711611319333315, 0.02701198123395443, -0.034915897995233536, 0.013866519555449486, 0.0019517126493155956, -0.010760419070720673, -0.027622107416391373, 0.04018517583608627, 0.011114015243947506, 0.020550182089209557, 0.01930219493806362, 0.0011977206449955702, 0.028925560414791107, 0.02247762866318226, 0.012001472525298595, -0.01308999489992857, 0.022546961903572083, -0.011613210663199425, -0.015807831659913063, 0.0034527634270489216, 0.006780728232115507, -0.005657540168613195, -0.013984384946525097, 0.022130966186523438, -0.027677573263645172, 0.009990827180445194, 0.025944259017705917, 0.06389692425727844, 0.002813170198351145, -0.0190664641559124, -0.012167870998382568, 0.004350620787590742, -0.00664552953094244, 0.02859276346862316, 0.023323485627770424, -0.017541147768497467, -0.022172564640641212, 0.014892642386257648, -0.008687375113368034, -0.011377479881048203, -0.009859095327556133, -0.018220607191324234, -0.009567898698151112, 0.00464528426527977, 0.021090976893901825, -0.021368306130170822, 0.027594374492764473, 0.02626318857073784, -0.0040074242278933525, 0.0047042169608175755, -0.002320908708497882, -0.016598224639892578, 0.004506618715822697, 0.019537925720214844, 0.016736889258027077, -0.0006192267755977809, -0.031643398106098175, 0.02365628257393837, 0.016764622181653976, -0.021271241828799248, -0.014025984331965446, -0.0017766478704288602, 0.003868758911266923, -0.005432208999991417, -0.015377970412373543, 0.011259614489972591, 0.007342322263866663, -0.006420198827981949, -0.008666574954986572, -0.02156243845820427, -0.02580559253692627, -0.005546608008444309, -0.010628687217831612, -0.009921494871377945, -0.026485051959753036, -0.02692878060042858]}, "text_id_to_ref_doc_id": {"2d707a34-28f5-4cfa-8973-961c2f7e96e1": "e62f594c-6592-44d1-aeca-eaf9ce7b7d9b"}, "metadata_dict": {"2d707a34-28f5-4cfa-8973-961c2f7e96e1": {"page_label": "1", "file_name": "ChatGPT\u4ea7\u54c1\u56e2\u961f 20230429.pdf", "_node_type": "TextNode", "document_id": "e62f594c-6592-44d1-aeca-eaf9ce7b7d9b", "doc_id": "e62f594c-6592-44d1-aeca-eaf9ce7b7d9b", "ref_doc_id": "e62f594c-6592-44d1-aeca-eaf9ce7b7d9b"}}}

joeshi/docstore.json

@@ -0,0 +1 @@
+{"docstore/metadata": {"e62f594c-6592-44d1-aeca-eaf9ce7b7d9b": {"doc_hash": "24d0fd2f1e041001a18d765f980998dd01a3587f8732ba0c139b4a7d782f5ed8"}, "2d707a34-28f5-4cfa-8973-961c2f7e96e1": {"doc_hash": "046d7c5337fef37cd7e84f404abfe6f83d7751a277b71725847539c993818f2c", "ref_doc_id": "e62f594c-6592-44d1-aeca-eaf9ce7b7d9b"}}, "docstore/data": {"2d707a34-28f5-4cfa-8973-961c2f7e96e1": {"__data__": {"id_": "2d707a34-28f5-4cfa-8973-961c2f7e96e1", "embedding": null, "metadata": {"page_label": "1", "file_name": "ChatGPT\u4ea7\u54c1\u56e2\u961f 20230429.pdf"}, "excluded_embed_metadata_keys": [], "excluded_llm_metadata_keys": [], "relationships": {"1": {"node_id": "e62f594c-6592-44d1-aeca-eaf9ce7b7d9b", "node_type": "4", "metadata": {"page_label": "1", "file_name": "ChatGPT\u4ea7\u54c1\u56e2\u961f 20230429.pdf"}, "hash": "24d0fd2f1e041001a18d765f980998dd01a3587f8732ba0c139b4a7d782f5ed8", "class_name": "RelatedNodeInfo"}}, "hash": "046d7c5337fef37cd7e84f404abfe6f83d7751a277b71725847539c993818f2c", "text": "\u79c1\u6709 ChatGPT\u4ea7\u54c1\u56e2\u961f\u4ecb\u7ecd\n1\u4ea7\u54c1\u53d1\u8d77\u4eba\uff0c\u9996\u5e2d\u5f00\u53d1\u4e13\u5bb6\uff1a\n\u77f3\u4e91\u535a\u58eb\uff0c\u5ba2\u6237\u4e16\u754c\u526f\u603b\u88c1\uff0c \u4e0a\u6d77\u8d22\u7ecf\u5927\u5b66\u91d1\u878d\u79d1\u6280\u7814\u7a76\u9662\u9ad8\u7ea7\u7814\n\u7a76\u5458\uff0c\u590d\u65e6\u5927\u5b66 \u8ba1\u7b97\u673a\u8f6f\u4ef6\u5b66\u9662\u5ba2\u5ea7\u8bb2\u5e08\n\u804c\u8d23\uff1a\n\u2022\u57fa\u4e8e ChatGPT\u5e94\u7528\u7684\u529f\u80fd\u7b97\u6cd5\u8bbe\u8ba1\u4e0e\u4f18\u5316\n\u2022\u4e2d\u6587\u573a\u666f\u5e94\u7528\u8bbe\u8ba1\u4e0e\u5f00\u53d1\n\u2022\u5398\u6e05\u5168\u7403\u53ca\u4e2d\u56fd\u56fd\u5185\u5927\u8bed\u97f3\u6a21\u578b\u6280\u672f\u8def\u7ebf\u56fe\n\u2022\u7ade\u54c1\u6d4b\u8bd5\u4e0e\u53cd\u9988\n\u6838\u5fc3\u5f00\u53d1\u56e2\u961f\u6210\u5458 \uff1a\n\u6881\u8d3a\u541b\u535a\u58eb\uff0c\u4e0a\u6d77\u8d22\u7ecf\u5927\u5b66\u91d1\u878d\u79d1\u6280\u7814\u7a76\u9662\u9ad8\u7ea7\u7814\u7a76 \u5458\nShirley Zhou \uff0c\u5fae\u8f6f\u5927\u4e2d\u534e\u533a\u4e91\u8ba1\u7b97\u5e73\u53f0 Azure\u9ad8\u7ea7\u6280\u672f\u4e3b\u7ba1\n\u5468\u5fd7\u82ac\uff0c\u67d0\u56fd\u5bb6\u7ea7\u4f01\u4e1a\u6570\u636e\u5e93\u5f00\u53d1\u8d1f\u8d23\u4eba\n\u804c\u8d23\uff1a\n\u2022\u57fa\u4e8e ChatGPT\u5e94\u7528\u5f00\u53d1\u4e0e\u5728\u7ebf \u90e8\u7f72\n\u2022\u4e2d\u6587\u573a\u666f\u5e94\u7528\u5f00\u53d1\u4e0e \u6d4b\u8bd5\n\u2022\u4ea7\u54c1\u8fd0\u7ef4\u6307\u5bfc\u6559\u6388 \uff1a\n\u97e9\u666f\u501c\u6559\u6388\uff0c\u535a\u58eb\u751f\u5bfc\u5e08\uff0c\u4e0a\u6d77\u8d22\u7ecf\u5927\u5b66\u4fe1\u606f\u5de5\u7a0b\u4e0e\u7ba1\u7406\u5b66\u9662\u5e38\u52a1\u526f\u9662\n\u957f\uff0c\u4e0a\u6d77\u8d22\u7ecf\u5927\u5b66\u91d1\u878d\u79d1\u6280\u7814\u7a76\u9662\u9662\u957f\n\u804c\u8d23\uff1a\n1.\u51b3\u5b9a\u5927\u8bed\u97f3\u6a21\u578b\u9879\u76ee \u7684\u6218\u7565\u65b9\u5411\u548c\u76ee\u6807\n2.\u786e\u4fdd\u9879\u76ee\u8d44\u6e90\u7684\u6295\u5165\n3.\u76d1\u63a7\u9879\u76ee\u98ce\u9669\u548c\u98ce\u9669\u7ba1\u7406\u7b56\u7565\n\u5e02\u573a\u4e0e\u9500\u552e \uff1a\n\u9648\u753b\uff0c\u67d0 TikTok\u8fd0\u8425\u5e73\u53f0\u8d1f\u8d23\u4eba\nCarrie Cao \uff0cTikTok\u56fd\u9645\u5546\u57ce\u8fd0\u8425\u9ad8\u7ea7\u4e3b\u7ba1\n\u804c\u8d23\uff1a\n1.\u62d3\u5c55\u4e2a\u4eba\u5ba2\u6237\u6e20\u9053\u53ca\u4f01\u4e1a\u7528\u6237\n2.\u7ef4\u7cfb\u5ba2\u6237\u5173\u7cfb\n3.\u6536\u96c6\u7ade\u54c1\u4fe1\u606f", "start_char_idx": 0, "end_char_idx": 496, "text_template": "{metadata_str}\n\n{content}", "metadata_template": "{key}: {value}", "metadata_seperator": "\n", "class_name": "TextNode"}, "__type__": "1"}}, "docstore/ref_doc_info": {"e62f594c-6592-44d1-aeca-eaf9ce7b7d9b": {"node_ids": ["2d707a34-28f5-4cfa-8973-961c2f7e96e1"], "metadata": {"page_label": "1", "file_name": "ChatGPT\u4ea7\u54c1\u56e2\u961f 20230429.pdf"}}}}

joeshi/faiss_index/index.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b44e357294efc4964845bcf8fa9e7fa95cdbe72f89157475f04f95376209ed65
+size 4361

joeshi/graph_store.json

@@ -0,0 +1 @@
+{"graph_dict": {}}

joeshi/image__vector_store.json

@@ -0,0 +1 @@
+{"embedding_dict": {}, "text_id_to_ref_doc_id": {}, "metadata_dict": {}}

joeshi/index_store.json

@@ -0,0 +1 @@
+{"index_store/data": {"bdcb3bbb-bbec-44f5-af7d-52bc9f2b06e9": {"__type__": "vector_store", "__data__": "{\"index_id\": \"bdcb3bbb-bbec-44f5-af7d-52bc9f2b06e9\", \"summary\": null, \"nodes_dict\": {\"2d707a34-28f5-4cfa-8973-961c2f7e96e1\": \"2d707a34-28f5-4cfa-8973-961c2f7e96e1\"}, \"doc_id_dict\": {}, \"embeddings_dict\": {}}"}}}

joeshi/joeshi_upload.csv

@@ -0,0 +1,201 @@
+员工,组别,平均通话时长,话务量,平均等待时间,工时利用率
+1,乙,22.08,60,109.14,75%
+2,乙,24.59,126,39.53,57%
+3,乙,12.0,141,45.65,72%
+4,甲,17.82,504,0.66,105%
+5,甲,16.95,394,106.61,129%
+6,丙,23.55,228,12.13,96%
+7,甲,21.35,53,24.68,37%
+8,乙,18.89,309,11.65,70%
+9,甲,22.05,349,17.11,88%
+10,丙,10.0,45,77.1,54%
+11,丙,44.27,38,69.13,62%
+12,乙,9.55,482,103.73,93%
+13,甲,15.79,269,83.66,76%
+14,丙,27.89,172,23.67,81%
+15,甲,32.76,78,41.09,91%
+16,甲,20.79,290,38.82,91%
+17,乙,6.6,459,76.78,84%
+18,乙,70.01,898,3.07,766%
+19,甲,10.0,814,143.95,128%
+20,甲,25.91,327,35.12,95%
+21,丙,14.75,358,11.46,78%
+22,甲,20.33,245,18.23,89%
+23,丙,3.2,239,157.58,70%
+24,甲,16.73,349,10.38,73%
+25,甲,13.23,143,130.79,87%
+26,甲,15.32,234,54.51,75%
+27,乙,16.98,388,26.27,81%
+28,乙,25.61,276,40.89,82%
+29,丙,26.68,188,26.67,84%
+30,丙,28.52,166,46.63,150%
+31,甲,23.29,274,10.49,81%
+32,丙,24.29,308,103.67,146%
+33,甲,20.46,305,60.78,86%
+34,丙,28.98,184,32.7,78%
+35,乙,27.07,248,23.51,86%
+36,甲,20.23,131,139.91,98%
+37,乙,15.6,370,3.4,78%
+38,乙,24.58,332,3.18,144%
+39,丙,14.24,605,2.69,87%
+40,乙,16.95,316,65.29,98%
+41,丙,25.83,301,22.51,94%
+42,丙,15.96,35,68.07,50%
+43,乙,13.03,33,51.33,44%
+44,丙,21.02,594,2.7,135%
+45,甲,28.45,379,102.45,144%
+46,丙,2.92,188,127.77,79%
+47,乙,12.95,314,45.43,76%
+48,甲,5.68,21,77.11,60%
+49,甲,4.65,10,142.83,85%
+50,甲,18.18,445,5.6,83%
+51,丙,21.99,137,66.46,73%
+52,甲,18.72,340,2.72,95%
+53,丙,17.53,108,7.4,78%
+54,丙,16.22,552,3.93,97%
+55,乙,16.31,40,14.46,17%
+56,甲,18.53,360,51.85,95%
+57,乙,12.21,445,2.89,96%
+58,丙,17.48,508,0.04,103%
+59,丙,6.68,111,117.19,77%
+60,丙,16.05,435,3.04,93%
+61,乙,23.31,456,4.26,96%
+62,丙,87.77,115,5.35,107%
+63,甲,30.08,251,0.11,143%
+64,乙,17.13,322,49.31,87%
+65,甲,22.55,322,8.95,81%
+66,乙,29.08,145,23.05,75%
+67,甲,91.11,130,10.04,123%
+68,甲,17.7,524,3.74,89%
+69,甲,24.58,258,19.14,82%
+70,丙,21.94,313,16.58,88%
+71,乙,21.9,172,47.2,82%
+72,甲,6.19,98,95.91,71%
+73,丙,10.8,173,73.72,74%
+74,甲,5.04,323,121.82,79%
+75,甲,17.42,580,2.34,101%
+76,乙,21.18,205,41.75,80%
+77,丙,12.34,494,4.77,69%
+78,丙,18.8,115,1.15,75%
+79,丙,12.38,36,66.28,49%
+80,乙,16.85,387,15.63,78%
+81,乙,18.86,188,6.97,79%
+82,甲,16.23,263,58.58,70%
+83,乙,11.3,40,0.0,6%
+84,丙,18.6,198,41.03,82%
+85,丙,21.75,224,12.23,69%
+86,乙,14.09,593,4.66,84%
+87,甲,12.07,109,83.17,74%
+88,丙,3.6,58,125.58,77%
+89,甲,19.7,477,41.59,100%
+90,丙,19.98,320,13.97,73%
+91,丙,24.65,283,18.44,81%
+92,甲,27.69,157,82.63,91%
+93,甲,22.84,84,153.37,64%
+94,丙,15.37,30,33.95,57%
+95,甲,30.75,109,2.53,57%
+96,乙,18.38,275,55.57,79%
+97,甲,18.1,373,67.18,110%
+98,丙,17.52,343,10.9,70%
+99,丙,19.06,355,94.92,126%
+100,丙,26.87,312,4.31,83%
+101,乙,2.71,87,118.05,66%
+102,乙,18.96,18,0.8,4%
+103,乙,16.51,574,3.75,101%
+104,甲,19.04,387,1.04,102%
+105,甲,19.05,346,28.32,82%
+106,丙,11.0,478,1.93,86%
+107,丙,8.11,281,26.4,38%
+108,乙,24.39,280,34.6,81%
+109,甲,37.29,211,24.79,87%
+110,乙,12.34,50,0.04,81%
+111,乙,22.27,227,58.91,133%
+112,乙,14.5,374,17.41,66%
+113,甲,17.82,406,27.11,107%
+114,乙,13.71,66,5.19,10%
+115,丙,19.38,37,65.41,56%
+116,甲,3.13,251,117.86,78%
+117,丙,15.32,122,98.46,78%
+118,甲,22.34,232,9.15,67%
+119,丙,17.42,400,16.82,79%
+120,甲,18.42,18,1.44,3%
+121,丙,28.16,197,29.18,76%
+122,乙,20.67,384,3.54,124%
+123,乙,7.12,379,100.1,82%
+124,丙,21.84,361,3.25,73%
+125,丙,9.97,10,84.59,85%
+126,丙,20.41,158,79.89,76%
+127,丙,25.0,246,11.21,67%
+128,丙,16.55,202,60.2,91%
+129,乙,18.76,452,1.41,117%
+130,丙,17.47,314,0.12,110%
+131,丙,15.63,51,13.85,70%
+132,甲,62.78,335,65.86,262%
+133,乙,21.59,223,86.08,116%
+134,甲,2.25,233,86.8,76%
+135,甲,19.72,229,86.59,91%
+136,甲,24.21,324,33.54,105%
+137,乙,14.0,335,66.23,85%
+138,丙,19.71,139,28.28,80%
+139,乙,22.08,195,9.73,65%
+140,丙,24.19,389,0.36,120%
+141,甲,3.58,51,134.34,75%
+142,丙,13.81,499,2.88,96%
+143,丙,11.34,352,77.42,86%
+144,丙,23.03,297,61.17,94%
+145,甲,15.22,595,6.47,84%
+146,丙,24.76,206,8.7,75%
+147,甲,2.6,194,144.26,81%
+148,丙,21.66,326,20.6,84%
+149,丙,12.08,479,105.67,92%
+150,甲,17.05,53,59.59,46%
+151,乙,20.21,423,31.86,98%
+152,丙,5.07,86,104.74,75%
+153,乙,15.2,648,3.1,97%
+154,丙,11.63,55,72.1,46%
+155,乙,28.5,46,0.68,73%
+156,乙,17.85,29,1.41,6%
+157,乙,24.62,293,89.46,138%
+158,甲,24.96,204,3.55,82%
+159,甲,10.95,951,2.89,105%
+160,丙,3.32,426,138.99,80%
+161,丙,7.45,502,95.35,98%
+162,甲,14.5,323,71.55,84%
+163,甲,18.99,301,85.58,116%
+164,甲,16.49,307,48.6,78%
+165,乙,15.95,281,44.09,67%
+166,甲,26.22,439,38.11,110%
+167,丙,17.18,119,30.92,42%
+168,丙,11.75,171,91.64,77%
+169,丙,16.79,374,18.87,80%
+170,乙,17.49,286,34.68,83%
+171,甲,20.15,343,15.98,78%
+172,丙,23.81,262,8.17,79%
+173,乙,27.47,230,7.05,77%
+174,乙,17.01,208,47.66,83%
+175,丙,20.8,390,22.55,114%
+176,丙,19.15,278,50.95,76%
+177,甲,16.76,271,56.42,75%
+178,甲,33.28,228,4.95,76%
+179,甲,24.57,427,62.71,114%
+180,丙,19.76,124,96.75,87%
+181,丙,2.85,36,158.6,83%
+182,甲,20.81,277,137.43,99%
+183,乙,16.07,312,111.49,96%
+184,丙,28.08,679,2.89,227%
+185,丙,14.78,488,3.89,79%
+186,丙,10.75,398,70.31,85%
+187,乙,16.25,385,10.97,84%
+188,甲,14.9,345,73.56,100%
+189,甲,27.54,309,37.5,105%
+190,丙,16.72,139,18.38,83%
+191,丙,28.82,172,46.19,81%
+192,乙,13.34,300,68.01,83%
+193,乙,9.98,401,68.94,82%
+194,乙,24.61,122,175.58,84%
+195,乙,4.34,357,103.76,76%
+196,乙,18.02,223,69.09,83%
+197,丙,32.28,53,67.35,59%
+198,甲,24.91,276,79.53,145%
+199,乙,18.57,201,41.61,78%
+200,丙,30.17,46,59.18,47%

joeshi/message.json

@@ -0,0 +1 @@
+"{\"data\":[{\"id\":\"msg_rAmEZriKWuh4ZOm3nm7mnSfR\",\"assistant_id\":\"asst_iyj1jY6FYX83y8UHKvD9GfGN\",\"content\":[{\"text\":{\"annotations\":[],\"value\":\"\u8ba9\u6211\u5148\u52a0\u8f7d\u6570\u636e\u96c6\uff0c\u7136\u540e\u5bf9\u6570\u636e\u8fdb\u884c\u521d\u6b65\u7684\u63a2\u7d22\u6027\u5206\u6790\u3002\u63a5\u7740\u6211\u5c06\u7ed8\u5236\u4e00\u4e9b\u6709\u8da3\u7684\u56fe\u8868\u6765\u5c55\u793a\u6570\u636e\u7684\u7279\u5f81\u548c\u8d8b\u52bf\u3002\u8ba9\u6211\u4eec\u5f00\u59cb\u5427\uff01\\n\\n\"},\"type\":\"text\"}],\"created_at\":1710314054,\"file_ids\":[],\"metadata\":{},\"object\":\"thread.message\",\"role\":\"assistant\",\"run_id\":\"run_mB2wrEzJBvI5ELOdk6lmezhW\",\"thread_id\":\"thread_cuANpHCi1VpxEnk381pEIh9g\"},{\"id\":\"msg_P8YumXlKMjs8375DUJljf089\",\"assistant_id\":null,\"content\":[{\"text\":{\"annotations\":[],\"value\":\"\u5206\u6790\u6b64\u6570\u636e\u96c6\u5e76\u7ed8\u5236\u4e00\u4e9b'\u6709\u8da3\u7684\u56fe\u8868'\uff0c\"},\"type\":\"text\"}],\"created_at\":1710314052,\"file_ids\":[\"file-Y0XiJYmVItTLPgmjfvK2TU3m\"],\"metadata\":{},\"object\":\"thread.message\",\"role\":\"user\",\"run_id\":null,\"thread_id\":\"thread_cuANpHCi1VpxEnk381pEIh9g\"}],\"object\":\"list\",\"first_id\":\"msg_rAmEZriKWuh4ZOm3nm7mnSfR\",\"last_id\":\"msg_P8YumXlKMjs8375DUJljf089\",\"has_more\":false}"

langchain_KB.py

@@ -0,0 +1,108 @@
+"""
+
+"""
+
+# -*- coding: utf-8 -*-
+import requests
+import streamlit as st
+import openai
+# from openai import embeddings
+import os
+from dotenv import load_dotenv
+import numpy as np
+import pandas as pd
+import csv
+import tempfile
+from tempfile import NamedTemporaryFile
+import pathlib
+from pathlib import Path
+import re
+from re import sub
+import matplotlib.pyplot as plt
+from itertools import product
+from tqdm import tqdm_notebook, tqdm, trange
+import time
+from time import sleep
+# import pretty_errors
+import seaborn as sns
+from matplotlib.pyplot import style
+from rich import print
+import warnings
+import PyPDF2
+from openai import OpenAI
+client = OpenAI()
+warnings.filterwarnings('ignore')
+
+''' 以下加载本地知识的核心内容。'''
+##! Install package, !pip install "unstructured[all-docs]", 需要完成这一步，否则会报错！
+# from langchain.document_loaders import UnstructuredFileLoader ## older version.
+from langchain_community.document_loaders.unstructured import UnstructuredFileLoader ## new version. 
+from langchain_community.document_loaders import PyPDFLoader
+
+from langchain.text_splitter import CharacterTextSplitter
+from langchain.embeddings.openai import OpenAIEmbeddings
+from langchain.embeddings.huggingface import HuggingFaceEmbeddings
+from langchain.vectorstores import FAISS
+
+load_dotenv()
+### 设置openai的API key
+os.environ["OPENAI_API_KEY"] = os.environ['user_token']
+openai.api_key = os.environ['user_token']
+
+# filepath = "/Users/yunshi/Downloads/txt_dir/Sparks_of_AGI.pdf"
+
+def langchain_localKB_construct(filepath, username):
+    print('开始构建Langchain知识库...')
+    # ''' 以下加载本地知识的核心内容。'''
+    ##! Install package, !pip install "unstructured[all-docs]", 需要完成这一步，否则会报错！
+
+    ## 加载文件
+    # filepath = "/Users/yunshi/Downloads/txt_dir/Sparks_of_AGI.pdf" ## a sample reference of a local PDF file. 
+    # loader = UnstructuredFileLoader(filepath)
+    
+    # from pathlib import Path
+    # filepath = Path(filepath)
+    print('now filepath:', filepath.name)
+    # loader = UnstructuredFileLoader(filepath.name) ### original code here.
+    loader = PyPDFLoader(filepath.name) ##NOTE: 只有PyPDFLoader才可以提取PDF的页数page信息。
+    # print('langchain loader:',loader)
+    docs = loader.load()
+    # print('docs now:', docs)
+
+    ## 文本分割
+    # text_splitter = CharacterTextSplitter(chunk_size=5000, chunk_overlap=200)
+    docs = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200).split_documents(docs)
+
+    ## 创建向量数据库
+    # embedding_model_name = 'GanymedeNil/text2vec-large-chinese'
+    # embeddings = HuggingFaceEmbeddings(model_name=embedding_model_name) ## 这里是联网情况下连接huggingface后使用。
+    from langchain.embeddings.openai import OpenAIEmbeddings
+    embeddings = OpenAIEmbeddings(disallowed_special=()) ## 可能需要更新了。
+    print('langchain embeddings:', embeddings)
+
+    vector_store = FAISS.from_documents(docs, embeddings)
+    # print(vector_store)
+    vector_store.save_local(f'./{username}/faiss_index')
+
+    return vector_store
+
+
+# vector_store = langchain_KB_construct(filepath='/Users/yunshi/Downloads/txt_dir/Sparks_of_AGI.pdf')
+# print(vs)
+
+
+
+
+### 根据prompt来检索本地知识库并回答。
+def langchain_RAG(prompt, username):
+    ### 用langchain看框架接入本地知识库。   
+    embeddings = OpenAIEmbeddings(disallowed_special=()) ## load embedding model again here. 
+    vector_store = FAISS.load_local(f'./{username}/faiss_index', embeddings)
+    docs = vector_store.similarity_search(prompt, k=5)
+    context = [doc.page_content for doc in docs]
+    total_prompt = f"已知信息：\n{context}\n 根据这些已知信息来回答问题:\n{prompt}"
+    # print('total prompt in local KB version:', total_prompt)
+
+    return total_prompt, docs
+
+# langchain_RAG('what are main challenges of AGI?')

localKB_construct.py

@@ -12,8 +12,8 @@ from llama_index import LLMPredictor
 from llama_index import ServiceContext
 from langchain.chat_models import ChatOpenAI
 from langchain import OpenAI
-from fastapi import FastAPI #* 实现流式数据
-from fastapi.responses import StreamingResponse #* 实现流式数据
+# from fastapi import FastAPI #* 实现流式数据
+# from fastapi.responses import StreamingResponse #* 实现流式数据
 import sys
 import os
 import math
@@ -32,8 +32,8 @@ import os
 from rich import print
 
 ## enironment settings. 
-os.environ["OPENAI_API_KEY"] = os.environ['user_token']
-openai.api_key = os.environ['user_token']
+os.environ["OPENAI_API_KEY"] = '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'
+openai.api_key = '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'
 # file_path = "/Users/yunshi/Downloads/txt_dir/Sparks_of_AGI.pdf"
 # file_path = "/Users/yunshi/Downloads/txt_dir/2023年百人会电动论坛 纪要 20230401.pdf"
 
@@ -65,9 +65,7 @@ def construct_index(directory_path):
     
     ## 好像work了，2023.09.22, 注意这里的写法有调整。
     # prompt_helper = PromptHelper(max_input_s≈ize, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)
-    # prompt_helper = PromptHelper(max_input_size, num_outputs, chunk_overlap_ratio= 0.1, chunk_size_limit=chunk_size_limit)
-    prompt_helper = PromptHelper(max_input_size=max_input_size, max_chunk_overlap=-1000, num_output=num_outputs, chunk_size_limit=chunk_size_limit)
-    
+    prompt_helper = PromptHelper(max_input_size, num_outputs, chunk_overlap_ratio= 0.1, chunk_size_limit=chunk_size_limit)
     service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
 
     ## 如果是txt文件，那么需要用如下命令。注意与PDF文件的区别。

message.json

@@ -0,0 +1 @@
+"{\"data\":[{\"id\":\"msg_kjVmuosPYiXK9xihJTek7G8l\",\"assistant_id\":\"asst_2o85VMcY0Sq1yj243C1GC36K\",\"content\":[{\"text\":{\"annotations\":[],\"value\":\"### 5.1 \u5206\u6790\u7ed3\u679c\\n\u4ece\u5c0f\u63d0\u7434\u56fe\u4e2d\u53ef\u4ee5\u770b\u51fa\uff0c\u8bdd\u52a1\u91cf\u7684\u5206\u5e03\u76f8\u5bf9\u96c6\u4e2d\u5728\u67d0\u4e2a\u8303\u56f4\u5185\uff0c\u540c\u65f6\u4e5f\u5b58\u5728\u4e00\u4e9b\u79bb\u7fa4\u503c\uff08Outlier\uff09\u3002\u5927\u90e8\u5206\u5458\u5de5\u7684\u8bdd\u52a1\u91cf\u5206\u5e03\u5728\u4e2d\u7b49\u6c34\u5e73\uff0c\u4f46\u4e5f\u6709\u4e00\u4e9b\u5458\u5de5\u7684\u8bdd\u52a1\u91cf\u8f83\u9ad8\u6216\u8f83\u4f4e\u3002\\n\\n### 5.2 \u5206\u6790\u539f\u56e0\\n\u53ef\u80fd\u9020\u6210\u8bdd\u52a1\u91cf\u5206\u5e03\u7279\u70b9\u7684\u539f\u56e0\u6709\uff1a\\n- \u5458\u5de5\u4e2a\u4eba\u80fd\u529b\u548c\u5de5\u4f5c\u6548\u7387\u4e0d\u540c\uff0c\u5bfc\u81f4\u8bdd\u52a1\u91cf\u5b58\u5728\u8f83\u5927\u7684\u5dee\u5f02\u3002\\n- \u4e0d\u540c\u7ec4\u522b\u7684\u5de5\u4f5c\u6027\u8d28\u548c\u4efb\u52a1\u5206\u914d\u4e0d\u540c\uff0c\u4e5f\u53ef\u80fd\u5bfc\u81f4\u8bdd\u52a1\u91cf\u7684\u5dee\u5f02\u6027\u3002\"},\"type\":\"text\"}],\"created_at\":1710316665,\"file_ids\":[],\"metadata\":{},\"object\":\"thread.message\",\"role\":\"assistant\",\"run_id\":\"run_04pRX93iX4UHdgS8dIc8niZG\",\"thread_id\":\"thread_uHcXOmtseqIunnLVR5cbPaK2\"},{\"id\":\"msg_otQ5IyeVU91Zr6JAu1IJVdov\",\"assistant_id\":\"asst_2o85VMcY0Sq1yj243C1GC36K\",\"content\":[{\"image_file\":{\"file_id\":\"file-ipbJKn14b9QNPqUGryEnTd4w\"},\"type\":\"image_file\"},{\"text\":{\"annotations\":[],\"value\":\"\u4ee5\u4e0a\u662f\u5173\u4e8e\u8bdd\u52a1\u91cf\u6570\u636e\u7684\u5c0f\u63d0\u7434\u56fe\u3002\u63a5\u4e0b\u6765\u6211\u5c06\u8fdb\u884c\u5206\u6790\u5e76\u63d0\u4f9b\u76f8\u5173\u7ed3\u679c\u3002\"},\"type\":\"text\"}],\"created_at\":1710316664,\"file_ids\":[],\"metadata\":{},\"object\":\"thread.message\",\"role\":\"assistant\",\"run_id\":\"run_04pRX93iX4UHdgS8dIc8niZG\",\"thread_id\":\"thread_uHcXOmtseqIunnLVR5cbPaK2\"},{\"id\":\"msg_8IEuLsMcJGCWSNbPN27yMq5x\",\"assistant_id\":null,\"content\":[{\"text\":{\"annotations\":[],\"value\":\"\u5bf9\u4e8e[\u8bdd\u52a1\u91cf]\u6570\u636e\u7ed9\u6211\u4e00\u4e2a'\u5c0f\u63d0\u7434\u56fe'\uff0c\u5e76\u7ed9\u51fa\u5206\u6790\u7ed3\u679c\u3002\"},\"type\":\"text\"}],\"created_at\":1710316648,\"file_ids\":[\"file-I2slYV5SIs4hGEwB6PGi9xzT\"],\"metadata\":{},\"object\":\"thread.message\",\"role\":\"user\",\"run_id\":null,\"thread_id\":\"thread_uHcXOmtseqIunnLVR5cbPaK2\"}],\"object\":\"list\",\"first_id\":\"msg_kjVmuosPYiXK9xihJTek7G8l\",\"last_id\":\"msg_8IEuLsMcJGCWSNbPN27yMq5x\",\"has_more\":false}"

multiQuery_prompt.py

@@ -0,0 +1,95 @@
+"""
+1. 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。
+
+
+"""
+
+# # from langchain.document_loaders import UnstructuredFileLoader
+# print('start the importing of UnstructuredFileLoader')
+# from langchain_community.document_loaders.unstructured import UnstructuredFileLoader
+# filepath = "./joeshi_upload.pdf"
+# loader = UnstructuredFileLoader(filepath)
+# docs = loader.load()
+# print('docs now:', docs)
+# print('langchain loader:',loader)
+# print('end the importing of UnstructuredFileLoader')
+from openai import OpenAI
+import openai
+import os
+import rag_source
+import langchain_KB
+from langchain.vectorstores import FAISS
+from langchain.embeddings.openai import OpenAIEmbeddings
+from langchain_community.document_loaders.unstructured import UnstructuredFileLoader ## new version. 
+from langchain_community.document_loaders import PyPDFLoader
+from langchain.text_splitter import CharacterTextSplitter
+from langchain.embeddings.openai import OpenAIEmbeddings
+from langchain.embeddings.huggingface import HuggingFaceEmbeddings
+from langchain.vectorstores import FAISS
+from dotenv import load_dotenv
+
+client = OpenAI()
+load_dotenv()
+### 设置openai的API key
+os.environ["OPENAI_API_KEY"] = os.environ['user_token']
+openai.api_key = os.environ['user_token']
+
+# loader = UnstructuredFileLoader(filepath)
+# loader = PyPDFLoader(filepath)
+# print('langchain loader:',loader)
+# docs = loader.load()
+# print('docs now:', docs)
+
+## 文本分割
+# text_splitter = CharacterTextSplitter(chunk_size=5000, chunk_overlap=200)
+# docs = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200).split_documents(docs)
+
+## 创建向量数据库
+# embedding_model_name = 'GanymedeNil/text2vec-large-chinese'
+# embeddings = HuggingFaceEmbeddings(model_name=embedding_model_name) ## 这里是联网情况下连接huggingface后使用。
+# from langchain.embeddings.openai import OpenAIEmbeddings
+# embeddings = OpenAIEmbeddings(disallowed_special=()) ## 可能需要更新了。
+# print('langchain embeddings:', embeddings)
+# vector_store = FAISS.from_documents(docs, embeddings)
+# print(vector_store)
+# vector_store.save_local('./joeshi/faiss_index') ## 将矢量库保存在本地指定目录。
+# vector_store = FAISS.load_local('./joeshi/faiss_index', embeddings)
+
+
+### 构建一个multi query的方法来进行大模型查询。可以用chatglm的方法来做, 也可以用openai的方法来做。
+def generate_sim_query(orignal_question):
+    similar_query_prompt = f"""你是一个AI语言模型，你的任务是生成4个不同的用户问题版本，以便从向量数据库中检索相关文档。通过生成用户问题多角度的表述，你的目标是帮助用户克服基于距离的相似性搜索的一些局限性。请按以下要求生成替代问题，并用隔行排列这些：
+    1. 使用与原始问题不同的关键词或同义词来重新表述问题。
+    2. 尝试从不同的角度提问，例如更具体、更抽象或更概括。
+    3. 考虑问题中隐含的前提或假设，并提出针对这些前提或假设的替代问题。
+    4. 结合用户可能拥有的背景知识或先前交互的信息，生成更具针对性的问题。
+    用户的原始问题是: {orignal_question}"""
+    # similar_query_prompt = """You are an AI language model assistant. Your task is to generate five 
+    # different versions of the given user question to retrieve relevant documents from a vector 
+    # database. By generating multiple perspectives on the user question, your goal is to help
+    # the user overcome some of the limitations of the distance-based similarity search. 
+    # Provide these alternative questions separated by newlines.
+    # Original question: {orignal_question}""" ## English version.
+    
+    #### 用大模型来生成相似问。
+    # response, history = chatglm.model.chat(chatglm.tokenizer, query=similar_query_prompt) ## 从用langchain的自定义方式来做.
+    response = client.chat.completions.create(
+                    model="gpt-3.5-turbo-16k",
+                    messages=[{"role": "user", "content": f'{similar_query_prompt}'}],
+                    stream=False,
+                )
+    
+    ## response.choices[0].message.content 才是ChatGPT反馈的返回的str结果。
+    similar_questions = response.choices[0].message.content.split("\n") ## 将反馈结果切割成一个个问题。
+    
+    return similar_questions
+
+
+# #### 根据相似问查询，返回查询后的结果以及信息来源。
+## 在主程序中会进行大模型的查询，这里应该不需要了。
+# def multiQuery_prompt(prompt):
+#     prompt = generate_sim_query(prompt)
+#     docs = vector_store.similarity_search(prompt, k=3) ##NOTE：注意可能新版本这里的k是小写的，之前的版本可能是大写。
+#     source = rag_source.rag_source(docs) ## get the K reference source of the RAG answer, in a designed format. 
+    
+#     return docs, source

rag_source.py

@@ -0,0 +1,68 @@
+"""
+1. 只有使用PyPDFLoader（from langchain_community.document_loaders import PyPDFLoader）, 才能获得在metadata中获得page的信息。
+
+"""
+
+from langchain_community.vectorstores import FAISS
+from langchain.embeddings.huggingface import HuggingFaceEmbeddings
+import streamlit as st
+import re
+from langchain.llms.base import LLM
+from langchain.llms.utils import enforce_stop_tokens
+from typing import Dict, List, Optional, Tuple, Union
+import requests
+import json
+
+# embeddings = HuggingFaceEmbeddings(model_name='/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/RAG/bge-large-zh/') ## 切换成BGE的embedding。
+
+# from langchain.embeddings.openai import OpenAIEmbeddings
+# embeddings = OpenAIEmbeddings(disallowed_special=()) ## 可能需要更新了。
+# vector_store = FAISS.load_local("./faiss_index/", embeddings=embeddings) ## 加载vector store到本地。
+
+## 在绝对路径中提取完整的文件名
+def extract_document_name(path):
+    # 路径分割
+    path_segments = path.split("/")
+    # 文件名提取
+    document_name = path_segments[-1]
+    return document_name
+
+## 从一段话中提取 1 句完整的句子，且该句子的长度必须超过 5 个词，同时去除了换行符'\n\n'。
+import re
+def extract_sentence(text):
+    """
+    从一段话中提取 1 句完整的句子，且该句子的长度必须超过 5 个词。
+
+    Args:
+        text: 一段话。
+
+    Returns:
+        提取到的句子。
+    """
+
+    # 去除换行符。
+    text = text.replace('\n\n', '')
+    # 使用正则表达式匹配句子。
+    sentences = re.split(r'[。？！；]', text)
+
+    # 过滤掉长度小于 5 个词的句子。
+    sentences = [sentence for sentence in sentences if len(sentence.split()) >= 5]
+
+    # 返回第一句句子。
+    return sentences[0] if sentences else None
+
+### 综合source的输出内容。
+##NOTE: 只有使用PyPDFLoader（from langchain_community.document_loaders import PyPDFLoader）, 才能获得在metadata中获得page的信息。
+def rag_source(docs):
+    print('starting source function!')
+    print('docs now:', docs)
+    source = ""
+    for i, doc in enumerate(docs):
+    # for i, doc in enumerate(docs): ### original code here.
+        source += f"**【信息来源 {i+1}】** " + extract_document_name(doc.metadata['source']) + '，' + f"第{docs[i].metadata['page']+1}页" + '，部分内容摘录：' + extract_sentence(doc.page_content) + '\n\n'
+        # source += f"**【信息来源 {i+1}】** " + extract_document_name(doc.metadata['source']) + '，' + f"第{docs[i].metadata['page']+1}页" + '，部分内容摘录：' + extract_sentence(doc.page_content) + '\n\n' 
+        ### original code here.
+    print('source:', source)
+    return source
+
+

save_database_info.py

@@ -10,9 +10,8 @@ from tqdm import tqdm_notebook, tqdm, trange
 import time
 import seaborn as sns
 from matplotlib.pyplot import style
-from rich import print
 import warnings
-warnings.filterwarnings('ignore')
+from datetime import datetime
 sns.set()
 # style.use('seaborn')
 

st_msautogen.py

@@ -0,0 +1,171 @@
+##TODO: 1. 多个图片处理。 2. 目前有个bug，在数据模式下，需要点击一下‘清除所有记录‘这个按键，或者重新输入一下内容。
+
+# -*- coding: utf-8 -*-
+import requests
+import streamlit as st
+import openai
+import os
+import datetime
+from dotenv import load_dotenv
+import numpy as np
+import pandas as pd
+import csv
+import tempfile
+from tempfile import NamedTemporaryFile
+import pathlib
+from pathlib import Path
+import re
+from re import sub
+import matplotlib.pyplot as plt
+from itertools import product
+from tqdm import tqdm_notebook, tqdm, trange
+import time
+from time import sleep
+import seaborn as sns
+from matplotlib.pyplot import style
+from rich import print
+import autogen
+from autogen import AssistantAgent
+from autogen import AssistantAgent, UserProxyAgent, config_list_from_json
+from st_reset_conversation import reset_all, reset_message
+import asyncio
+# import file
+# import pyqt5
+from PIL import Image  # type: ignore
+import warnings
+warnings.filterwarnings('ignore')
+# sns.set()
+import matplotlib
+matplotlib.use('Agg') ## https://blog.51cto.com/u_15642578/5301647
+
+load_dotenv()
+### 设置openai的API key
+os.environ["OPENAI_API_KEY"] = os.environ['user_token']
+openai.api_key = os.environ['user_token']
+
+# reset_all()
+
+# st.title('Microsoft AutoGen - Streamlit Demo')
+
+##NOTE： working，可以在MS AutoGen中的agent的system_message中添加一段提示信息。而且不会在st中显示。
+time_stamp = str(time.strftime("%Y-%m-%d %H:%M:%S", time.localtime())) + str(np.random.randint(0, 1000000))
+
+## 以下两个类基本相同，唯一的区别是基类不同。
+class TrackableAssistentAgent(AssistantAgent):
+    def _process_received_message(self, message, sender, silent):
+        with st.chat_message(sender.name): ##sender.name好像是streamlit的要求。
+            if "execution succeeded" in message:
+                if os.path.exists(f'/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Microsoft AutoGen/output_{time_stamp}.png'):
+                    st.image(f'/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Microsoft AutoGen/output_{time_stamp}.png')
+                    st.markdown(message)
+                    st.markdown('任务已经完成，请查看屏幕上的显示信息！')
+                else:
+                    st.markdown(message)
+                    st.markdown('任务已经完成，请查看屏幕上的显示信息！')
+                st.stop()
+            else: 
+                st.markdown(message)
+            
+        return super()._process_received_message(message, sender, silent)
+
+## 以下两个类基本相同，唯一的区别是基类不同。
+class TrackableUserProxyAgent(UserProxyAgent):
+    def _process_received_message(self, message, sender, silent):
+        # with st.status('思考中...', expanded=True, state='running') as status:
+        with st.chat_message(sender.name):
+            if "execution succeeded" in message:
+                if os.path.exists(f'/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Microsoft AutoGen/output_{time_stamp}.png'):
+                    # st.markdown('有图片！！')
+                    st.image(f'/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Microsoft AutoGen/output_{time_stamp}.png')
+                    st.markdown(message)
+                    st.markdown('任务已经完成，请查看屏幕上的显示信息！')
+                else:
+                    st.markdown(message)
+                    st.markdown('任务已经完成，请查看屏幕上的显示信息！')
+                st.stop()
+            else:
+                st.markdown(message)
+        return super()._process_received_message(message, sender, silent)
+
+
+
+
+# async def auto_gen(uploaded_file_path): ## async一定需要在主程序中用asyncio.run(st_msautogen.auto_gen(uploaded_file_path))启动。
+def auto_gen(uploaded_file_path):
+    # st.title('Microsoft AutoGen - Streamlit Demo')
+
+    ##TODO 是否需要如下代码？
+    ### 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"])
+    # print('msautogen uploaded_file_path:', uploaded_file_path)
+    print('running time:', datetime.datetime.now().strftime("%H:%M:%S.%f"))
+
+    prompt = st.chat_input("说点什么吧...")
+    # print('prompt:', prompt)
+
+    system_prompt = f"""You are a helpful AI assistant. Solve tasks using your coding and language skills. In the following cases, suggest python code (in a python coding block) or shell script (in a sh coding block) for the user to execute.
+    1. When you need to collect info, use the code to output the info you need, for example, browse or search the web, download/read a file, print the content of a webpage or a file, get the current date/time, check the operating system. After sufficient info is printed and the task is ready to be solved based on your language skill, you can solve the task by yourself.
+    2. When you need to perform some task with code, use the code to perform the task and output the result. Finish the task smartly. Solve the task step by step if you need to. If a plan is not provided, explain your plan first. Be clear which step uses code, and which step uses your language skill.When using code, you must indicate the script type in the code block. The user cannot provide any other feedback or perform any other action beyond executing the code you suggest. The user can't modify your code. So do not suggest incomplete code which requires users to modify. Don't use a code block if it's not intended to be executed by the user. If you want the user to save the code in a file before executing it, put # filename: <filename> inside the code block as the first line. Don't include multiple code blocks in one response. Do not ask users to copy and paste the result. Instead, use 'print' function for the output when relevant. Check the execution result returned by the user.If the result indicates there is an error, fix the error and output the code again. Suggest the full code instead of partial code or code changes. If the error can't be fixed or if the task is not solved even after the code is executed successfully, analyze the problem, revisit your assumption, collect additional info you need, and think of a different approach to try. When you find an answer, verify the answer carefully. Include verifiable evidence in your response if possible. 
+    Reply "TERMINATE" in the end when everything is done.
+    Other settings include:
+    * 如果我需要你分析上传的文件，那么文件的位置在：{uploaded_file_path}
+    * 如果要求你输出图表，那么图的解析度dpi需要设定为300。图使用seaborn库。seaborn库的参数设定：'axes.facecolor':'#FFF9ED','figure.facecolor':'#FFF9ED', palette='dark'。
+    * 如果需要你输出图片，那么将图片保持到以下地址：'/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Microsoft AutoGen/output_{time_stamp}.png'。
+    * 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：
+        1. 提供详细且专业的分析结果，提供足够的分析依据。
+        2. 给出可能造成这一结果的可能原因有哪些？
+        以上这些内容全部用序列号格式来表达。
+    """
+    
+    if prompt:
+        llm_config = {
+            # "request_timeout": 600,
+            "config_list":[{
+                "model": "gpt-4-1106-preview",
+                # "model": "gpt-4",
+                # "model": "gpt-3.5-turbo-16k",
+                "api_key": os.environ["OPENAI_API_KEY"],
+            }], ## 注意这里的格式，否则会在后期输出会报错
+        }
+
+        ## check if the prompt has been passed in. 
+        print('prompt after llm_config:', prompt)
+        
+        ## create an assistant.
+        assistant = TrackableAssistentAgent(name='assistant', llm_config=llm_config, system_message=system_prompt)
+
+        ## create a user proxy. 
+        ### see notes at https://microsoft.github.io/autogen/docs/FAQ#use-the-constructed-configuration-list-in-agents
+        user_proxy = TrackableUserProxyAgent(name='user',human_input_mode='NEVER',llm_config=llm_config)
+
+        ## create a loop event
+        loop = asyncio.new_event_loop()
+        # Set the loop as the event loop.
+        asyncio.set_event_loop(loop)
+
+        ### define an async function.
+        async def start_chat():
+            try:
+                # print('333')
+                await user_proxy.a_initiate_chat(assistant, message=prompt, silent=True)
+                # user_proxy.a_initiate_chat(assistant, message=prompt, silent=True) ## not working, the function not even start up. 
+                print('444')
+            except Exception as e:
+                print('start_chat function not working because:', e)
+                pass
+        
+        print('111')
+        ## run the async function within the event loop.
+        loop.run_until_complete(start_chat())
+        print('222')
+        
+        # # #TODO：尝试停止一个async中的loop。
+        loop.close()
+
+# auto_gen()

st_openai_assistant.py

@@ -0,0 +1,291 @@
+'''
+使用OpenAI的官方Assistant，实现code_interpreter功能:
+1. 使用OpenAI官方的API教程： 
+  https://cookbook.openai.com/examples/assistants_api_overview_python （重要）
+  https://platform.openai.com/docs/assistants/overview
+2. YT视频： https://www.youtube.com/watch?v=5rcjGjgJNQc
+3. 基础代码参考：openai_interpreter.py
+4. 函数化text_response和image_response和prompt。
+5. 中文字体显示还是有问题，现在妥协的方式是转换成英文来显示。不影响分析过程和结果。
+
+##TEMP
+
+        import matplotlib
+        import seaborn as sns
+        SimHei = FontProperties(fname='SimHei.ttf')
+        matplotlib.rcParams['font.sans-serif'] = ['SimHei']
+        sns.set(rc={'axes.facecolor':'#FFF9ED','figure.facecolor':'#FFF9ED'}, palette='dark', font='SimHei')
+        fontproperties = SimHei
+
+每次输出图表时，你都需要在matplotlib和seaborn中安装中文字体'SimHei'，安装的路径在：'/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/SimHei.ttf'。然后，你需要在mattplotlib和seaborn中设置字体为'SimHei'。
+
+'''
+##TODO:
+
+# -*- coding: utf-8 -*-
+import csv
+import os
+import pathlib
+import re
+import tempfile
+import time
+import warnings
+from itertools import product
+from pathlib import Path
+from re import sub
+from tempfile import NamedTemporaryFile
+from time import sleep
+
+import matplotlib.pyplot as plt
+import numpy as np
+import openai
+import pandas as pd
+import requests
+import seaborn as sns
+from dotenv import load_dotenv
+from matplotlib.pyplot import style
+from rich import print
+from tqdm import tqdm, tqdm_notebook, trange
+import chatgpt ## 自定义的chatgpt函数，用于优化text response的回答。
+
+import streamlit as st
+import add_fonts
+
+warnings.filterwarnings('ignore')
+sns.set()
+
+load_dotenv()
+### 设置openai的API key
+os.environ["OPENAI_API_KEY"] = os.environ['user_token']
+openai.api_key = os.environ['user_token']
+
+from openai import Client
+client = Client()
+
+## JSON output display.   
+import json
+def show_json(name=None,obj=None):
+    # display(json.loads(obj.model_dump_json())) ## orignial code.
+    print(name,":",json.loads(obj.model_dump_json()))
+    print("--"*40) 
+
+def save_json(obj):
+    json_message = json.dumps(obj.model_dump_json(), indent=4)
+    with open('message.json', 'w') as file:
+        file.write(json_message)
+
+### set up a function to wrap up the code-interpreter.
+def openai_assistant(prompt=None, filepath=None, username=None):
+    client = Client()
+    file = client.files.create(
+        file=open(filepath, 'rb'),
+        # file=open('./大模型LLM解决方案调研问卷.pdf', 'rb'), 
+        purpose='assistants',
+    )
+    # show_json(file) ## view the file info. 
+
+    ### create an ansistant. 
+    assistant = client.beta.assistants.create(
+    name="AI Expert",
+    # instructions="You are a helpful and powerful AI. When asked a question, you need to answer the question based on the information in the file provided to you.",
+    instructions="""
+    你是一个强大的AI助手。当被问到一个问题时，你需要根据提供给你的文件中的信息来回答这个问题。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。
+    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。
+    3. 图表上如果有非英文的文字，那么你需要将字体翻译为英文，然后显示。
+    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。
+    """,
+    # tools=[{"type": "retrieval"}],
+    tools=[{"type": "code_interpreter"}],
+    # model="gpt-4-1106-preview",
+    model="gpt-3.5-turbo-1106", ## 注意这个版本以上才有retrieval功能。
+    file_ids=[file.id],
+    # file_ids=["file-6ZaFecYrnYDFXH7WM3HbtQRK"], ## what are the OpenAI assistant API's file_ids look like. 
+    )
+    # print('assistant ID:', assistant.id)
+    # show_json(assistant)
+
+    ### 这里必须要提供file.id。
+    thread = client.beta.threads.create(
+        messages=[
+        {
+            "role": "user",
+            # "content": "analyze the dataset and plot something interesting.",
+            "content": prompt,
+            # "file_ids": ["file-6ZaFecYrnYDFXH7WM3HbtQRK"],
+            "file_ids": [file.id]
+        }
+        ],
+    )
+    run = client.beta.threads.runs.create(
+        thread_id = thread.id,
+        assistant_id = assistant.id,
+    )
+
+    run = client.beta.threads.runs.retrieve(
+        run_id = run.id,
+        thread_id = thread.id,
+        timeout=100,
+    )
+
+    ##NOTE: 因为run是异步的，所以这里必须要等待一段时间，直到run完成。否则返回的结果不会有内容，只有user的prompt。
+    import time
+    def wait_on_run(run, thread):
+        while run.status == "queued" or run.status == "in_progress":
+            run = client.beta.threads.runs.retrieve(
+                thread_id=thread.id,
+                run_id=run.id,
+            )
+            time.sleep(0.5)
+        return run
+    run = wait_on_run(run, thread)
+
+    ### retrieve the message from OpenAI. 
+    messages = client.beta.threads.messages.list(thread_id=thread.id) ##NOTE: 注意这里的返回格式默认是倒序，可以设置成正序。但是最后一个往往是我们要的答案。
+    
+    show_json(name='messages:',obj=messages)
+    print('--'*40)
+    save_json(obj=messages)
+
+    ## check the messages format. find image and context accordingly. 
+    # text_response, image_response = None, None ## single image and text response.
+    
+    ###NOTE: find all the image files inside the reponsed message.
+    image_response = [] ## multiple images and text responses.
+    imagefile_count = 0
+    imagefile_position = []
+    for i in range(len(messages.data)):
+        for j in range(len(messages.data[i].content)):
+            try:
+                if messages.data[i].content[j].image_file:
+                    imagefile_count += 1
+                    imagefile_position.append((i,j))
+            except:
+                pass
+
+            print('--'*30)
+            print("总共有几张图片？：", imagefile_count)
+            print('--'*30)
+                
+    print('start the image and text repsonse process!')
+    
+    image_name = []
+    image_files = []
+    for x, y in imagefile_position:
+        random_num = np.random.randint(10000, 50000)
+        print('x,y=', x, y)
+        try:
+            if messages.data[x].content[y].image_file:
+                ### save the image file.
+                image_file = openai.files.content(file_id=messages.data[x].content[y].image_file.file_id) ## message's image file id. note the format. 
+                image_files.append(image_file)
+                
+                ##NOTE: 这里会保存图片在本地目录中，考虑关闭。
+                # with open(f'./{username}/output{x}{y}_{random_num}.png', "wb") as f:
+                #     f.write(image_file.content)
+
+                # # Load the PNG file 
+                # import matplotlib.image as mpimg
+                # import matplotlib.pyplot as plt
+                # # image_response.append(mpimg.imread(f'./{username}/output{x}{y}.png'))
+                # image_response = mpimg.imread(f'./{username}/output{x}{y}_{random_num}.png')
+                # plt.imshow(image_response)
+                # plt.show()
+        except Exception as e:
+            print(f"An error occurred: {e}")
+            pass
+        
+    ### consolidate text_response below. 
+    text_response_num = 0
+    my_msg = []
+    for x in range(5): #NOTE: 遍历所有的数据点，但是注意要去掉最后的一段。
+        for y in range(5):
+            try:
+                if messages.data[x].content[y].text:
+                    print('x, y=', x, y)
+                    my_msg.append(messages.data[x].content[y].text.value)
+                    text_response_num += 1
+            except:
+                pass
+    final_msg = sorted(my_msg[1:], reverse=True) ## 需要去除最后两段，因为是默认的user的输入，还有就是第一个无用的assistant接受命令的回答。
+    text_response = str()
+    for i in range(len(final_msg)):
+        text_response += final_msg[i]
+    print('final_msg:', final_msg)
+    print('总共有几个text response：', text_response_num)
+    
+    ## convert a readiable markdown format.
+    # text_response = final_msg[0] #NOTE：输出的格式是list，需要转换成Markdown可以是识别的内容。这里是获得一个str格式的内容，否则一个list对象。
+    # text_response = text_response.replace("['",'"""').replace("']",'"""') ## 这里不需要处理首尾的list符号。
+    
+    ## 用LLM优化所有的text response的回答。
+    import chatgpt
+    # text_response = '中国在哪里？'
+    user_prompt = f"""首先，我会向你提供一段【文字内容】，这段文字中可能包括了一系列的多轮对话的内容。接着，我需要你根据这段文字中的内容整理成一段文字结论。你的回答风格需要很专业，包括：尽可能的包含统计数据、数字和专业的结论，不能有口语化的表达。【文字内容】如下{text_response}。"""
+    final_answer = chatgpt.chatgpt(user_prompt=user_prompt)
+    # final_answer = chatgpt.chatgpt(user_prompt=user_prompt, openai_model="gpt-3.5-turbo-16k") ### original code here.
+    
+    return messages, text_response, image_response, image_files, final_answer
+
+
+# filepath = '/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/rawdata/iris.csv'
+# # # # # # # prompt = "analyze the dataset and plot something interesting."
+
+# prompt = "analyze the dataset and plot something interesting. give me at least 2 plots. please reply in Chinese instead of English."
+# # # # # # # messages, text_reponse, image_response = openai_assistant(filepath=filepath,username='joeshi') ## working.
+# messages, text_response, image_response, image_files, final_answer = openai_assistant(prompt, filepath=filepath,username='joeshi') ## probable multiple images and text responses.
+
+# print("--"*40)
+# print('final_answer:', final_answer)
+# print("--"*40)
+
+
+# # # '''temp work area'''
+# # # # msg_json = show_json(messages)
+# # show_json(name='messages:',obj=messages)
+
+# # # messages.data[0].content[0].image_file
+# # # messages.data[0].content[2].text.value
+
+# # # tt = text_response
+
+# # # import re
+# # # regex = re.compile(r'(\n\n)|(\n1)')
+# # # def convert_markdown(text):
+# # #     """
+# # #     将markdown格式的文字转换成可读性强的文字格式。
+# # #     Args:
+# # #         text: markdown格式的文字。
+# # #     Returns:
+# # #         可读性强的文字。
+# # #     """
+# # #     result = re.sub(regex, lambda m: m.group(1) if m.group(1) else '\n1.', text)
+# # #     return result
+# # # print(convert_markdown(str(text_response)))
+
+# # # messages.data[1].content[1].text.value
+
+# my_msg = []
+# for x in range(5):
+#     for y in range(5):
+#         try:
+#             if messages.data[x].content[y].text:
+#                 print('x, y=', x, y)
+#                 my_msg.append(messages.data[x].content[y].text.value)
+#             # else:
+#             #     continue
+#         except:
+#             pass
+# print(sorted(my_msg[:-1], reverse=True))
+
+# msg = str()
+# for i in range(len(final_msg)):
+#     msg += final_msg[i]
+    
+# print(msg)
+# type(msg)

st_reset_conversation.py

@@ -0,0 +1,15 @@
+## reset all conversion inside streamlit
+## clear conversion.
+import streamlit as st
+def reset_all():
+    # st.session_state.conversation = None ## 关闭这个可以保持当前的模型选项？
+    st.session_state.chat_history = None
+    st.session_state.messages = []
+    message_placeholder = st.empty()
+    return None
+
+
+def reset_message():
+    st.session_state.messages = []
+    message_placeholder = st.empty()
+    return None