Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,174 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from dotenv import load_dotenv
|
3 |
+
from PyPDF2 import PdfReader
|
4 |
+
from langchain.text_splitter import CharacterTextSplitter, RecursiveCharacterTextSplitter
|
5 |
+
from langchain.embeddings import OpenAIEmbeddings, HuggingFaceInstructEmbeddings
|
6 |
+
from langchain.vectorstores import FAISS, Chroma
|
7 |
+
from langchain.embeddings import HuggingFaceEmbeddings # General embeddings from HuggingFace models.
|
8 |
+
from langchain.chat_models import ChatOpenAI
|
9 |
+
from langchain.memory import ConversationBufferMemory
|
10 |
+
from langchain.chains import ConversationalRetrievalChain
|
11 |
+
from htmlTemplates import css, bot_template, user_template
|
12 |
+
from langchain.llms import HuggingFaceHub, LlamaCpp, CTransformers # For loading transformer models.
|
13 |
+
from langchain.document_loaders import PyPDFLoader, TextLoader, JSONLoader, CSVLoader
|
14 |
+
import tempfile # ์์ ํ์ผ์ ์์ฑํ๊ธฐ ์ํ ๋ผ์ด๋ธ๋ฌ๋ฆฌ์
๋๋ค.
|
15 |
+
import os
|
16 |
+
|
17 |
+
|
18 |
+
# PDF ๋ฌธ์๋ก๋ถํฐ ํ
์คํธ๋ฅผ ์ถ์ถํ๋ ํจ์์
๋๋ค.
|
19 |
+
def get_pdf_text(pdf_docs):
|
20 |
+
temp_dir = tempfile.TemporaryDirectory() # ์์ ๋๋ ํ ๋ฆฌ๋ฅผ ์์ฑํฉ๋๋ค.
|
21 |
+
temp_filepath = os.path.join(temp_dir.name, pdf_docs.name) # ์์ ํ์ผ ๊ฒฝ๋ก๋ฅผ ์์ฑํฉ๋๋ค.
|
22 |
+
with open(temp_filepath, "wb") as f: # ์์ ํ์ผ์ ๋ฐ์ด๋๋ฆฌ ์ฐ๊ธฐ ๋ชจ๋๋ก ์ฝ๋๋ค.
|
23 |
+
f.write(pdf_docs.getvalue()) # PDF ๋ฌธ์์ ๋ด์ฉ์ ์์ ํ์ผ์ ์๋๋ค.
|
24 |
+
pdf_loader = PyPDFLoader(temp_filepath) # PyPDFLoader๋ฅผ ์ฌ์ฉํด PDF๋ฅผ ๋ก๋ํฉ๋๋ค.
|
25 |
+
pdf_doc = pdf_loader.load() # ํ
์คํธ๋ฅผ ์ถ์ถํฉ๋๋ค.
|
26 |
+
return pdf_doc # ์ถ์ถํ ํ
์คํธ๋ฅผ ๋ฐํํฉ๋๋ค.
|
27 |
+
|
28 |
+
# ๊ณผ์
|
29 |
+
# ์๋ ํ
์คํธ ์ถ์ถ ํจ์๋ฅผ ์์ฑ
|
30 |
+
|
31 |
+
def get_text_file(text_docs):
|
32 |
+
temp_dir2 = tempfile.TemporaryDirectory() # ์์ ๋๋ ํ ๋ฆฌ๋ฅผ ์์ฑํฉ๋๋ค.
|
33 |
+
temp_filepath2 = os.path.join(temp_dir2.name, text_docs.name) # ์์ ํ์ผ ๊ฒฝ๋ก๋ฅผ ์์ฑํฉ๋๋ค.
|
34 |
+
with open(temp_filepath2, "wb") as f: # ์์ ํ์ผ์ ๋ฐ์ด๋๋ฆฌ ์ฐ๊ธฐ ๋ชจ๋๋ก ์ฝ๋๋ค.
|
35 |
+
f.write(text_docs.getvalue()) # text ๋ฌธ์์ ๋ด์ฉ์ ์์ ํ์ผ์ ์๋๋ค.
|
36 |
+
text_loader = TextLoader(temp_filepath2) # TextLoader๋ฅผ ์ฌ์ฉํด text๋ฅผ ๋ก๋ํฉ๋๋ค.
|
37 |
+
text_doc = text_loader.load() # ํ
์คํธ๋ฅผ ์ถ์ถํฉ๋๋ค.
|
38 |
+
return text_doc # ์ถ์ถํ ํ
์คํธ๋ฅผ ๋ฐํํฉ๋๋ค.
|
39 |
+
|
40 |
+
def get_csv_file(csv_docs):
|
41 |
+
temp_dir3 = tempfile.TemporaryDirectory() # ์์ ๋๋ ํ ๋ฆฌ๋ฅผ ์์ฑํฉ๋๋ค.
|
42 |
+
temp_filepath3 = os.path.join(temp_dir3.name, csv_docs.name) # ์์ ํ์ผ ๊ฒฝ๋ก๋ฅผ ์์ฑํฉ๋๋ค.
|
43 |
+
with open(temp_filepath3, "wb") as f: # ์์ ํ์ผ์ ๋ฐ์ด๋๋ฆฌ ์ฐ๊ธฐ ๋ชจ๋๋ก ์ฝ๋๋ค.
|
44 |
+
f.write(csv_docs.getvalue()) # csv ๋ฌธ์์ ๋ด์ฉ์ ์์ ํ์ผ์ ์๋๋ค.
|
45 |
+
csv_loader = CSVLoader(temp_filepath3) # CSVLoader๋ฅผ ์ฌ์ฉํด csv๋ฅผ ๋ก๋ํฉ๋๋ค.
|
46 |
+
csv_doc = csv_loader.load() # ํ
์คํธ๋ฅผ ์ถ์ถํฉ๋๋ค.
|
47 |
+
return csv_doc # ์ถ์ถํ ํ
์คํธ๋ฅผ ๋ฐํํฉ๋๋ค.
|
48 |
+
|
49 |
+
def get_json_file(json_docs):
|
50 |
+
temp_dir4 = tempfile.TemporaryDirectory() # ์์ ๋๋ ํ ๋ฆฌ๋ฅผ ์์ฑํฉ๋๋ค.
|
51 |
+
temp_filepath4 = os.path.join(temp_dir4.name, json_docs.name) # ์์ ํ์ผ ๊ฒฝ๋ก๋ฅผ ์์ฑํฉ๋๋ค.
|
52 |
+
with open(temp_filepath4, "wb") as f: # ์์ ํ์ผ์ ๋ฐ์ด๋๋ฆฌ ์ฐ๊ธฐ ๋ชจ๋๋ก ์ฝ๋๋ค.
|
53 |
+
f.write(json_docs.getvalue()) # json ๋ฌธ์์ ๋ด์ฉ์ ์์ ํ์ผ์ ์๋๋ค.
|
54 |
+
json_loader = JSONLoader(temp_filepath4, jq_schema=<your_json_schema>) # JSONLoader๋ฅผ ์ฌ์ฉํด json๋ฅผ ๋ก๋ํฉ๋๋ค.
|
55 |
+
json_doc = json_loader.load() # ํ
์คํธ๋ฅผ ์ถ์ถํฉ๋๋ค.
|
56 |
+
return json_doc # ์ถ์ถํ ํ
์คํธ๋ฅผ ๋ฐํํฉ๋๋ค.
|
57 |
+
|
58 |
+
|
59 |
+
# ๋ฌธ์๋ค์ ์ฒ๋ฆฌํ์ฌ ํ
์คํธ ์ฒญํฌ๋ก ๋๋๋ ํจ์์
๋๋ค.
|
60 |
+
def get_text_chunks(documents):
|
61 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
62 |
+
chunk_size=1000, # ์ฒญํฌ์ ํฌ๊ธฐ๋ฅผ ์ง์ ํฉ๋๋ค.
|
63 |
+
chunk_overlap=200, # ์ฒญํฌ ์ฌ์ด์ ์ค๋ณต์ ์ง์ ํฉ๋๋ค.
|
64 |
+
length_function=len # ํ
์คํธ์ ๊ธธ์ด๋ฅผ ์ธก์ ํ๋ ํจ์๋ฅผ ์ง์ ํฉ๋๋ค.
|
65 |
+
)
|
66 |
+
|
67 |
+
documents = text_splitter.split_documents(documents) # ๋ฌธ์๋ค์ ์ฒญํฌ๋ก ๋๋๋๋ค
|
68 |
+
if not documents:
|
69 |
+
raise ValueError("์
๋ก๋๋ ๋ฌธ์๊ฐ ์๊ฑฐ๋ ๋ชจ๋ ๋ฌธ์๊ฐ ๋น์ด ์์ต๋๋ค.")
|
70 |
+
return documents # ๋๋ ์ฒญํฌ๋ฅผ ๋ฐํํฉ๋๋ค.
|
71 |
+
|
72 |
+
|
73 |
+
# ํ
์คํธ ์ฒญํฌ๋ค๋ก๋ถํฐ ๋ฒกํฐ ์คํ ์ด๋ฅผ ์์ฑํ๋ ํจ์์
๋๋ค.
|
74 |
+
def get_vectorstore(text_chunks):
|
75 |
+
# OpenAI ์๋ฒ ๋ฉ ๋ชจ๋ธ์ ๋ก๋ํฉ๋๋ค. (Embedding models - Ada v2)
|
76 |
+
|
77 |
+
embeddings = OpenAIEmbeddings()
|
78 |
+
vectorstore = FAISS.from_documents(text_chunks, embeddings) # FAISS ๋ฒกํฐ ์คํ ์ด๋ฅผ ์์ฑํฉ๋๋ค.
|
79 |
+
|
80 |
+
return vectorstore # ์์ฑ๋ ๋ฒกํฐ ์คํ ์ด๋ฅผ ๋ฐํํฉ๋๋ค.
|
81 |
+
|
82 |
+
|
83 |
+
def get_conversation_chain(vectorstore):
|
84 |
+
gpt_model_name = 'gpt-3.5-turbo'
|
85 |
+
llm = ChatOpenAI(model_name = gpt_model_name) #gpt-3.5 ๋ชจ๋ธ ๋ก๋
|
86 |
+
|
87 |
+
# ๋ํ ๊ธฐ๋ก์ ์ ์ฅํ๊ธฐ ์ํ ๋ฉ๋ชจ๋ฆฌ๋ฅผ ์์ฑํฉ๋๋ค.
|
88 |
+
memory = ConversationBufferMemory(
|
89 |
+
memory_key='chat_history', return_messages=True)
|
90 |
+
# ๋ํ ๊ฒ์ ์ฒด์ธ์ ์์ฑํฉ๋๋ค.
|
91 |
+
conversation_chain = ConversationalRetrievalChain.from_llm(
|
92 |
+
llm=llm,
|
93 |
+
retriever=vectorstore.as_retriever(),
|
94 |
+
memory=memory
|
95 |
+
)
|
96 |
+
return conversation_chain
|
97 |
+
|
98 |
+
# ์ฌ์ฉ์ ์
๋ ฅ์ ์ฒ๋ฆฌํ๋ ํจ์์
๋๋ค.
|
99 |
+
def handle_userinput(user_question):
|
100 |
+
# ๋ํ ์ฒด์ธ์ ์ฌ์ฉํ์ฌ ์ฌ์ฉ์ ์ง๋ฌธ์ ๋ํ ์๋ต์ ์์ฑํฉ๋๋ค.
|
101 |
+
response = st.session_state.conversation({'question': user_question})
|
102 |
+
# ๋ํ ๊ธฐ๋ก์ ์ ์ฅํฉ๋๋ค.
|
103 |
+
st.session_state.chat_history = response['chat_history']
|
104 |
+
|
105 |
+
for i, message in enumerate(st.session_state.chat_history):
|
106 |
+
if i % 2 == 0:
|
107 |
+
st.write(user_template.replace(
|
108 |
+
"{{MSG}}", message.content), unsafe_allow_html=True)
|
109 |
+
else:
|
110 |
+
st.write(bot_template.replace(
|
111 |
+
"{{MSG}}", message.content), unsafe_allow_html=True)
|
112 |
+
|
113 |
+
|
114 |
+
def main():
|
115 |
+
load_dotenv()
|
116 |
+
st.set_page_config(page_title="Chat with multiple Files",
|
117 |
+
page_icon=":books:")
|
118 |
+
st.write(css, unsafe_allow_html=True)
|
119 |
+
|
120 |
+
if "conversation" not in st.session_state or st.session_state.conversation is None:
|
121 |
+
st.session_state.conversation = None
|
122 |
+
st.session_state.chat_history = None
|
123 |
+
|
124 |
+
st.header("Chat with multiple Files :")
|
125 |
+
user_question = st.text_input("Ask a question about your documents:")
|
126 |
+
|
127 |
+
# "Send" ๋ฒํผ
|
128 |
+
if st.button("Send"):
|
129 |
+
if user_question:
|
130 |
+
handle_userinput(user_question)
|
131 |
+
|
132 |
+
if user_question:
|
133 |
+
handle_userinput(user_question)
|
134 |
+
|
135 |
+
with st.sidebar:
|
136 |
+
openai_key = st.text_input("Paste your OpenAI API key (sk-...)")
|
137 |
+
if openai_key:
|
138 |
+
os.environ["OPENAI_API_KEY"] = openai_key
|
139 |
+
|
140 |
+
st.subheader("Your documents")
|
141 |
+
docs = st.file_uploader(
|
142 |
+
"Upload your PDFs, TEXTfiles, CSVfiles, JSONfiles here and click on 'Process'", accept_multiple_files=True)
|
143 |
+
if st.button("Process"):
|
144 |
+
with st.spinner("Processing"):
|
145 |
+
# get pdf text
|
146 |
+
doc_list = []
|
147 |
+
|
148 |
+
for file in docs:
|
149 |
+
print('file - type : ', file.type)
|
150 |
+
if file.type == 'text/plain':
|
151 |
+
# file is .txt
|
152 |
+
doc_list.extend(get_text_file(file))
|
153 |
+
elif file.type in ['application/octet-stream', 'application/pdf']:
|
154 |
+
# file is .pdf
|
155 |
+
doc_list.extend(get_pdf_text(file))
|
156 |
+
elif file.type == 'text/csv':
|
157 |
+
# file is .csv
|
158 |
+
doc_list.extend(get_csv_file(file))
|
159 |
+
elif file.type == 'application/json':
|
160 |
+
# file is .json
|
161 |
+
doc_list.extend(get_json_file(file))
|
162 |
+
|
163 |
+
# get the text chunks
|
164 |
+
text_chunks = get_text_chunks(doc_list)
|
165 |
+
|
166 |
+
# create vector store
|
167 |
+
vectorstore = get_vectorstore(text_chunks)
|
168 |
+
|
169 |
+
# create conversation chain
|
170 |
+
st.session_state.conversation = get_conversation_chain(
|
171 |
+
vectorstore)
|
172 |
+
|
173 |
+
if __name__ == '__main__':
|
174 |
+
main()
|