File size: 7,799 Bytes
bd2cf7b
e221fec
 
 
b8484c0
bd2cf7b
 
e221fec
93ac82b
bd2cf7b
 
 
 
8421d14
9359dde
 
bd2cf7b
 
 
 
 
 
 
 
 
 
 
 
 
 
8421d14
 
58f8dd3
8421d14
 
 
 
 
 
 
 
 
 
 
 
bd2cf7b
8421d14
 
 
bd2cf7b
8421d14
 
 
 
7508414
8421d14
 
5ef51e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ea780f0
b8484c0
 
 
 
 
 
 
affd12c
b8484c0
 
 
 
 
affd12c
b8484c0
 
 
 
 
 
 
 
affd12c
b8484c0
affd12c
 
 
 
 
b8484c0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8421d14
 
 
e221fec
 
 
 
b8484c0
 
93ac82b
 
 
b8484c0
 
 
 
 
e221fec
8421d14
 
 
 
5ef51e2
 
9359dde
 
5ef51e2
 
9359dde
 
bd2cf7b
e221fec
 
 
 
 
 
 
 
 
 
5ef51e2
e221fec
b8484c0
 
 
 
8421d14
b8484c0
 
5ef51e2
e221fec
 
 
 
 
 
5ef51e2
e221fec
b8484c0
5ef51e2
 
b8484c0
 
5ef51e2
b8484c0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
import streamlit as st
import pandas as pd
import numpy as np
import datetime
import gspread
import pickle
import os
import csv
import json
import torch
from tqdm.auto import tqdm
from langchain.text_splitter import RecursiveCharacterTextSplitter


# from langchain.vectorstores import Chroma
from langchain.vectorstores import FAISS
from langchain.embeddings import HuggingFaceInstructEmbeddings


from langchain import HuggingFacePipeline
from langchain.chains import RetrievalQA



st.set_page_config(
    page_title = 'aitGPT',
    page_icon = 'βœ…')



@st.cache_data
def load_scraped_web_info():
    with open("ait-web-document", "rb") as fp:
        ait_web_documents = pickle.load(fp)
        
        
    text_splitter = RecursiveCharacterTextSplitter(
        # Set a really small chunk size, just to show.
        chunk_size = 500,
        chunk_overlap  = 100,
        length_function = len,
    )

    chunked_text = text_splitter.create_documents([doc for doc in tqdm(ait_web_documents)])


@st.cache_resource
def load_embedding_model():
    embedding_model = HuggingFaceInstructEmbeddings(model_name='hkunlp/instructor-base',
                                                model_kwargs = {'device': torch.device('cuda' if torch.cuda.is_available() else 'cpu')})
    return embedding_model

@st.cache_data
def load_faiss_index():
    vector_database = FAISS.load_local("faiss_index_web_and_curri_new", embedding_model) #CHANGE THIS FAISS EMBEDDED KNOWLEDGE
    return vector_database

@st.cache_resource
def load_llm_model():
    # llm = HuggingFacePipeline.from_model_id(model_id= 'lmsys/fastchat-t5-3b-v1.0', 
    #                                         task= 'text2text-generation',
    #                                         model_kwargs={ "device_map": "auto",
    #                                                     "load_in_8bit": True,"max_length": 256, "temperature": 0,
    #                                                     "repetition_penalty": 1.5})
    
    
    llm = HuggingFacePipeline.from_model_id(model_id= 'lmsys/fastchat-t5-3b-v1.0', 
                                        task= 'text2text-generation',
                                        
                                        model_kwargs={ "max_length": 256, "temperature": 0,
                                                      "torch_dtype":torch.float32,
                                                    "repetition_penalty": 1.3})
    return llm


def load_retriever(llm, db):
    qa_retriever = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff",
                            retriever=db.as_retriever())

    return qa_retriever

def retrieve_document(query_input):
    related_doc = vector_database.similarity_search(query_input)
    return related_doc

def retrieve_answer(query_input):
    prompt_answer=  query_input + " " + "Try to elaborate as much as you can."
    answer = qa_retriever.run(prompt_answer)
    output = st.text_area(label="Retrieved documents", value=answer[6:]) #this positional slicing helps remove "<pad> " at the beginning
    
    st.markdown('---')
    score = st.radio(label = 'please select the rating score for overall satifaction and helpfullness of the bot answer', options=[0, 1,2,3,4,5], horizontal=True,
                     on_change=update_worksheet_qa, key='rating')

    return answer[6:] #this positional slicing helps remove "<pad> " at the beginning
    
# def update_score():
#     st.session_state.session_rating = st.session_state.rating


def update_worksheet_qa():
    st.session_state.session_rating = st.session_state.rating
    #This if helps validate the initiated rating, if 0, then the google sheet would not be updated
    #(edited) now even with the score of 0, we still want to store the log because some users do not give the score to complete the logging
    if st.session_state.session_rating  == 0:
        worksheet_qa.append_row([st.session_state.history[-1]['timestamp'].strftime(datetime_format), 
                                st.session_state.history[-1]['question'],
                                st.session_state.history[-1]['generated_answer'], 
                                st.session_state.session_rating 
                                ])
    else:
        worksheet_qa.append_row([st.session_state.history[-1]['timestamp'].strftime(datetime_format), 
                                st.session_state.history[-1]['question'],
                                st.session_state.history[-1]['generated_answer'], 
                                st.session_state.session_rating 
                                ])
        
def update_worksheet_comment():
    worksheet_comment.append_row([datetime.datetime.now().strftime(datetime_format),
                                feedback_input])
    success_message = st.success('Feedback successfully submitted, thank you', icon="βœ…",
               )
    time.sleep(3)
    success_message.empty()

#--------------


if "history" not in st.session_state:
    st.session_state.history = []
if "session_rating" not in st.session_state:
    st.session_state.session_rating = 0


credentials= json.loads(st.secrets['google_sheet_credential'])

service_account = gspread.service_account_from_dict(credentials)
workbook= service_account.open("aitGPT-qa-log")
worksheet_qa = workbook.worksheet("Sheet1")
worksheet_comment = workbook.worksheet("Sheet2")
datetime_format= "%Y-%m-%d %H:%M:%S"



load_scraped_web_info()
embedding_model = load_embedding_model()
vector_database = load_faiss_index()
llm_model = load_llm_model()
qa_retriever = load_retriever(llm= llm_model, db= vector_database)


print("all load done")




    



st.write("# aitGPT πŸ€– ")
st.markdown("""
         #### The aitGPT project is a virtual assistant developed by the :green[Asian Institute of Technology] that contains a vast amount of information gathered from 205 AIT-related websites.  
        The goal of this chatbot is to provide an alternative way for applicants and current students to access information about the institute, including admission procedures, campus facilities, and more.  
          """)
st.write(' ⚠️ Please expect to wait **~ 10 - 20 seconds per question** as thi app is running on CPU against 3-billion-parameter LLM')

st.markdown("---")
st.write(" ")
st.write("""
         ### ❔ Ask a question
         """)

query_input = st.text_area(label= 'What would you like to know about AIT?' , key = 'my_text_input')
generate_button = st.button(label = 'Ask question!')

if generate_button:
    answer = retrieve_answer(query_input)
    log = {"timestamp": datetime.datetime.now(),
        "question":query_input,
        "generated_answer": answer,
        "rating":st.session_state.session_rating }

    st.session_state.history.append(log)
    update_worksheet_qa()


st.write(" ")
st.write(" ")

st.markdown("---")
st.write("""
         ### πŸ’Œ Your voice matters
         """)

feedback_input = st.text_area(label= 'please leave your feedback or any ideas to make this bot more knowledgeable and fun')
feedback_button = st.button(label = 'Submit feedback!')

if feedback_button:
    update_worksheet_comment()


# if st.session_state.session_rating == 0:
#     pass
# else:
#     with open('test_db', 'a') as csvfile:
#         writer = csv.writer(csvfile)
#         writer.writerow([st.session_state.history[-1]['timestamp'], st.session_state.history[-1]['question'],
#                             st.session_state.history[-1]['generated_answer'], st.session_state.session_rating ])
#         st.session_state.session_rating = 0

# test_df = pd.read_csv("test_db", index_col=0)
# test_df.sort_values(by = ['timestamp'],
#                     axis=0,
#                     ascending=False,
#                     inplace=True)
# st.dataframe(test_df)