Spaces:
Paused
Paused
Commit
β’
f3517b6
1
Parent(s):
74d04be
Update pages/3_chat.py
Browse files- pages/3_chat.py +119 -13
pages/3_chat.py
CHANGED
@@ -1,7 +1,108 @@
|
|
1 |
import streamlit as st
|
2 |
-
from
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
from langchain.chains import RetrievalQA
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
with st.expander("key information"):
|
7 |
st.write( st.session_state['chunked_df'], unsafe_allow_html=True)
|
@@ -10,19 +111,24 @@ with st.expander("key information"):
|
|
10 |
st.markdown(st.session_state['repetition_penalty'])
|
11 |
|
12 |
|
13 |
-
# qa_retriever = RetrievalQA.from_chain_type(llm=st.session_state['llm_model'] , chain_type="stuff",
|
14 |
-
# retriever=st.session_state['faiss_db'].as_retriever())
|
15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
|
17 |
|
18 |
-
# with chat(key="my_chat"):
|
19 |
-
# if prompt := st.chat_input():
|
20 |
-
# add_message("user", prompt, avatar="π§βπ»")
|
21 |
-
# # def stream_echo():
|
22 |
-
# # for word in prompt.split():
|
23 |
-
# # yield word + " "
|
24 |
-
# # time.sleep(0.15)
|
25 |
-
# add_message("assistant", "Echo: ", qa_retriever.run(prompt), avatar="π¦")
|
26 |
|
27 |
-
|
28 |
-
|
|
|
1 |
import streamlit as st
|
2 |
+
from streamlit_chat import message as st_message
|
3 |
+
import pandas as pd
|
4 |
+
import numpy as np
|
5 |
+
import datetime
|
6 |
+
import gspread
|
7 |
+
import torch
|
8 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
9 |
+
|
10 |
+
|
11 |
+
# from langchain.vectorstores import Chroma
|
12 |
+
from langchain.vectorstores import FAISS
|
13 |
+
from langchain.embeddings import HuggingFaceInstructEmbeddings
|
14 |
+
|
15 |
+
|
16 |
+
from langchain import HuggingFacePipeline
|
17 |
from langchain.chains import RetrievalQA
|
18 |
+
from langchain.prompts import PromptTemplate
|
19 |
+
from langchain.memory import ConversationBufferWindowMemory
|
20 |
+
|
21 |
+
|
22 |
+
from langchain.chains import LLMChain
|
23 |
+
from langchain.chains import ConversationalRetrievalChain
|
24 |
+
from langchain.chains.question_answering import load_qa_chain
|
25 |
+
from langchain.chains.conversational_retrieval.prompts import CONDENSE_QUESTION_PROMPT
|
26 |
+
|
27 |
+
|
28 |
+
prompt_template = """
|
29 |
+
You are the chatbot and your job is to give answers.
|
30 |
+
MUST only use the following pieces of context to answer the question at the end. If the answers are not in the context or you are not sure of the answer, just say that you don't know, don't try to make up an answer.
|
31 |
+
{context}
|
32 |
+
Question: {question}
|
33 |
+
When encountering abusive, offensive, or harmful language, such as fuck, bitch,etc, just politely ask the users to maintain appropriate behaviours.
|
34 |
+
Always make sure to elaborate your response.
|
35 |
+
Never answer with any unfinished response
|
36 |
+
Answer:
|
37 |
+
"""
|
38 |
+
PROMPT = PromptTemplate(
|
39 |
+
template=prompt_template, input_variables=["context", "question"]
|
40 |
+
)
|
41 |
+
# chain_type_kwargs = {"prompt": PROMPT}
|
42 |
+
|
43 |
+
@st.cache_resource
|
44 |
+
def load_conversational_qa_memory_retriever(llm_model, vector_database):
|
45 |
+
|
46 |
+
question_generator = LLMChain(llm=llm_model, prompt=CONDENSE_QUESTION_PROMPT)
|
47 |
+
doc_chain = load_qa_chain(llm_model, chain_type="stuff", prompt = PROMPT)
|
48 |
+
memory = ConversationBufferWindowMemory(k = 3, memory_key="chat_history", return_messages=True, output_key='answer')
|
49 |
+
|
50 |
+
|
51 |
+
|
52 |
+
conversational_qa_memory_retriever = ConversationalRetrievalChain(
|
53 |
+
retriever=vector_database.as_retriever(),
|
54 |
+
question_generator=question_generator,
|
55 |
+
combine_docs_chain=doc_chain,
|
56 |
+
return_source_documents=True,
|
57 |
+
memory = memory,
|
58 |
+
get_chat_history=lambda h :h)
|
59 |
+
return conversational_qa_memory_retriever, question_generator
|
60 |
+
|
61 |
+
def new_retrieve_answer():
|
62 |
+
prompt_answer= st.session_state.my_text_input + ". Try to be elaborate and informative in your answer."
|
63 |
+
answer = conversational_qa_memory_retriever({"question": prompt_answer })
|
64 |
+
|
65 |
+
print(f"condensed quesion : {question_generator.run({'chat_history': answer['chat_history'], 'question' : prompt_answer})}")
|
66 |
+
|
67 |
+
print(answer["chat_history"])
|
68 |
+
|
69 |
+
st.session_state.chat_history.append({"message": st.session_state.my_text_input, "is_user": True})
|
70 |
+
st.session_state.chat_history.append({"message": answer['answer'][6:] , "is_user": False})
|
71 |
+
|
72 |
+
st.session_state.my_text_input = ""
|
73 |
+
|
74 |
+
return answer['answer'][6:] #this positional slicing helps remove "<pad> " at the beginning
|
75 |
+
|
76 |
+
def clean_chat_history():
|
77 |
+
st.session_state.chat_history = []
|
78 |
+
conversational_qa_memory_retriever.memory.chat_memory.clear() #add this to remove
|
79 |
+
|
80 |
+
|
81 |
+
if "history" not in st.session_state: #this one is for the google sheet logging
|
82 |
+
st.session_state.history = []
|
83 |
+
|
84 |
|
85 |
+
if "chat_history" not in st.session_state: #this one is to pass previous messages into chat flow
|
86 |
+
st.session_state.chat_history = []
|
87 |
+
|
88 |
+
|
89 |
+
|
90 |
+
|
91 |
+
conversational_qa_memory_retriever, question_generator = load_conversational_qa_memory_retriever(llm_model = st.session_state['model'],
|
92 |
+
vector_database = st.session_state['faiss_db'])
|
93 |
+
|
94 |
+
|
95 |
+
|
96 |
+
print("all load done")
|
97 |
+
|
98 |
+
|
99 |
+
# Try adding this to set to clear the memory in each session
|
100 |
+
if st.session_state.chat_history == []:
|
101 |
+
conversational_qa_memory_retriever.memory.chat_memory.clear()
|
102 |
+
|
103 |
+
|
104 |
+
|
105 |
+
st.write("# extraGPT π€ ")
|
106 |
|
107 |
with st.expander("key information"):
|
108 |
st.write( st.session_state['chunked_df'], unsafe_allow_html=True)
|
|
|
111 |
st.markdown(st.session_state['repetition_penalty'])
|
112 |
|
113 |
|
|
|
|
|
114 |
|
115 |
+
st.write(' β οΈ Please expect to wait **~ 10 - 20 seconds per question** as thi app is running on CPU against 3-billion-parameter LLM')
|
116 |
+
|
117 |
+
st.markdown("---")
|
118 |
+
st.write(" ")
|
119 |
+
st.write("""
|
120 |
+
### β Ask a question
|
121 |
+
""")
|
122 |
+
|
123 |
+
|
124 |
+
|
125 |
+
|
126 |
+
for chat in st.session_state.chat_history:
|
127 |
+
st_message(**chat)
|
128 |
+
|
129 |
+
query_input = st.text_input(label= 'Type a question' , key = 'my_text_input', on_change= new_retrieve_answer )
|
130 |
|
131 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
132 |
|
133 |
+
clear_button = st.button("Start new convo",
|
134 |
+
on_click=clean_chat_history)
|