wnsdud030415 commited on
Commit
7e634bd
ยท
1 Parent(s): d870d0c

Upload app.py

Browse files
Files changed (1) hide show
  1. app.py +177 -0
app.py ADDED
@@ -0,0 +1,177 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 # ์ž„์‹œ ํŒŒ์ผ์„ ์ƒ์„ฑํ•˜๊ธฐ ์œ„ํ•œ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ์ž…๋‹ˆ๋‹ค.
15
+ import os
16
+
17
+
18
+ # PDF ๋ฌธ์„œ๋กœ๋ถ€ํ„ฐ ํ…์ŠคํŠธ๋ฅผ ์ถ”์ถœํ•˜๋Š” ํ•จ์ˆ˜์ž…๋‹ˆ๋‹ค.
19
+ def get_pdf_text(pdf_docs):
20
+ temp_dir = tempfile.TemporaryDirectory() # ์ž„์‹œ ๋””๋ ‰ํ† ๋ฆฌ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
21
+ temp_filepath = os.path.join(temp_dir.name, pdf_docs.name) # ์ž„์‹œ ํŒŒ์ผ ๊ฒฝ๋กœ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
22
+ with open(temp_filepath, "wb") as f: # ์ž„์‹œ ํŒŒ์ผ์„ ๋ฐ”์ด๋„ˆ๋ฆฌ ์“ฐ๊ธฐ ๋ชจ๋“œ๋กœ ์—ฝ๋‹ˆ๋‹ค.
23
+ f.write(pdf_docs.getvalue()) # PDF ๋ฌธ์„œ์˜ ๋‚ด์šฉ์„ ์ž„์‹œ ํŒŒ์ผ์— ์”๋‹ˆ๋‹ค.
24
+ pdf_loader = PyPDFLoader(temp_filepath) # PyPDFLoader๋ฅผ ์‚ฌ์šฉํ•ด PDF๋ฅผ ๋กœ๋“œํ•ฉ๋‹ˆ๋‹ค.
25
+ pdf_doc = pdf_loader.load() # ํ…์ŠคํŠธ๋ฅผ ์ถ”์ถœํ•ฉ๋‹ˆ๋‹ค.
26
+ return pdf_doc # ์ถ”์ถœํ•œ ํ…์ŠคํŠธ๋ฅผ ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค.
27
+
28
+
29
+ # ๊ณผ์ œ
30
+ # ์•„๋ž˜ ํ…์ŠคํŠธ ์ถ”์ถœ ํ•จ์ˆ˜๋ฅผ ์ž‘์„ฑ
31
+
32
+
33
+ def get_text_file(docs):
34
+ temp_dir = tempfile.TemporaryDirectory() # ์ž„์‹œ ๋””๋ ‰ํ† ๋ฆฌ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
35
+ temp_filepath = os.path.join(temp_dir.name, docs.name) # ์ž„์‹œ ํŒŒ์ผ ๊ฒฝ๋กœ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
36
+ with open(temp_filepath, "wb") as f: # ์ž„์‹œ ํŒŒ์ผ์„ ๋ฐ”์ด๋„ˆ๋ฆฌ ์“ฐ๊ธฐ ๋ชจ๋“œ๋กœ ์—ฝ๋‹ˆ๋‹ค.
37
+ f.write(docs.getvalue()) # ๋ฌธ์„œ์˜ ๋‚ด์šฉ์„ ์ž„์‹œ ํŒŒ์ผ์— ์”๋‹ˆ๋‹ค.
38
+ loader = TextLoader(temp_filepath) # TextLoader๋ฅผ ์‚ฌ์šฉํ•ด text๋ฅผ ๋กœ๋“œํ•ฉ๋‹ˆ๋‹ค.
39
+ doc = loader.load() # ํ…์ŠคํŠธ๋ฅผ ์ถ”์ถœํ•ฉ๋‹ˆ๋‹ค.
40
+ return doc # ์ถ”์ถœํ•œ ํ…์ŠคํŠธ๋ฅผ ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค.
41
+
42
+
43
+ def get_csv_file(docs):
44
+ temp_dir = tempfile.TemporaryDirectory() # ์ž„์‹œ ๋””๋ ‰ํ† ๋ฆฌ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
45
+ temp_filepath = os.path.join(temp_dir.name, docs.name) # ์ž„์‹œ ํŒŒ์ผ ๊ฒฝ๋กœ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
46
+ with open(temp_filepath, "wb") as f: # ์ž„์‹œ ํŒŒ์ผ์„ ๋ฐ”์ด๋„ˆ๋ฆฌ ์“ฐ๊ธฐ ๋ชจ๋“œ๋กœ ์—ฝ๋‹ˆ๋‹ค.
47
+ f.write(docs.getvalue()) # ๋ฌธ์„œ์˜ ๋‚ด์šฉ์„ ์ž„์‹œ ํŒŒ์ผ์— ์”๋‹ˆ๋‹ค.
48
+ loader = CSVLoader(temp_filepath) # csvLoader๋ฅผ ์‚ฌ์šฉํ•ด csv๋ฅผ ๋กœ๋“œํ•ฉ๋‹ˆ๋‹ค.
49
+ doc = loader.load() # ํ…์ŠคํŠธ๋ฅผ ์ถ”์ถœํ•ฉ๋‹ˆ๋‹ค.
50
+ return doc # ์ถ”์ถœํ•œ ํ…์ŠคํŠธ๋ฅผ ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค.
51
+
52
+
53
+ def get_json_file(docs):
54
+ temp_dir = tempfile.TemporaryDirectory() # ์ž„์‹œ ๋””๋ ‰ํ† ๋ฆฌ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
55
+ temp_filepath = os.path.join(temp_dir.name, docs.name) # ์ž„์‹œ ํŒŒ์ผ ๊ฒฝ๋กœ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
56
+ with open(temp_filepath, "wb") as f:
57
+ f.write(docs.getvalue())
58
+ f.seek(0)
59
+ json_loader = JSONLoader(f.name,
60
+ jq_schema='.scans[].relationships',
61
+ text_content=False)
62
+ json_doc = json_loader.load()
63
+
64
+ return json_doc
65
+
66
+
67
+ # ๋ฌธ์„œ๋“ค์„ ์ฒ˜๋ฆฌํ•˜์—ฌ ํ…์ŠคํŠธ ์ฒญํฌ๋กœ ๋‚˜๋ˆ„๋Š” ํ•จ์ˆ˜์ž…๋‹ˆ๋‹ค.
68
+ def get_text_chunks(documents):
69
+ text_splitter = RecursiveCharacterTextSplitter(
70
+ chunk_size=1000, # ์ฒญํฌ์˜ ํฌ๊ธฐ๋ฅผ ์ง€์ •ํ•ฉ๋‹ˆ๋‹ค.
71
+ chunk_overlap=200, # ์ฒญํฌ ์‚ฌ์ด์˜ ์ค‘๋ณต์„ ์ง€์ •ํ•ฉ๋‹ˆ๋‹ค.
72
+ length_function=len # ํ…์ŠคํŠธ์˜ ๊ธธ์ด๋ฅผ ์ธก์ •ํ•˜๋Š” ํ•จ์ˆ˜๋ฅผ ์ง€์ •ํ•ฉ๋‹ˆ๋‹ค.
73
+ )
74
+
75
+ documents = text_splitter.split_documents(documents) # ๋ฌธ์„œ๋“ค์„ ์ฒญํฌ๋กœ ๋‚˜๋ˆ•๋‹ˆ๋‹ค
76
+ return documents # ๋‚˜๋ˆˆ ์ฒญํฌ๋ฅผ ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค.
77
+
78
+
79
+ # ํ…์ŠคํŠธ ์ฒญํฌ๋“ค๋กœ๋ถ€ํ„ฐ ๋ฒกํ„ฐ ์Šคํ† ์–ด๋ฅผ ์ƒ์„ฑํ•˜๋Š” ํ•จ์ˆ˜์ž…๋‹ˆ๋‹ค.
80
+ def get_vectorstore(text_chunks):
81
+ # OpenAI ์ž„๋ฒ ๋”ฉ ๋ชจ๋ธ์„ ๋กœ๋“œํ•ฉ๋‹ˆ๋‹ค. (Embedding models - Ada v2)
82
+
83
+ embeddings = OpenAIEmbeddings()
84
+ vectorstore = FAISS.from_documents(text_chunks, embeddings) # FAISS ๋ฒกํ„ฐ ์Šคํ† ์–ด๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
85
+
86
+ return vectorstore # ์ƒ์„ฑ๋œ ๋ฒกํ„ฐ ์Šคํ† ์–ด๋ฅผ ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค.
87
+
88
+
89
+ def get_conversation_chain(vectorstore):
90
+ gpt_model_name = 'gpt-3.5-turbo'
91
+ llm = ChatOpenAI(model_name=gpt_model_name) # gpt-3.5 ๋ชจ๋ธ ๋กœ๋“œ
92
+
93
+ # ๋Œ€ํ™” ๊ธฐ๋ก์„ ์ €์žฅํ•˜๊ธฐ ์œ„ํ•œ ๋ฉ”๋ชจ๋ฆฌ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
94
+ memory = ConversationBufferMemory(
95
+ memory_key='chat_history', return_messages=True)
96
+ # ๋Œ€ํ™” ๊ฒ€์ƒ‰ ์ฒด์ธ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
97
+ conversation_chain = ConversationalRetrievalChain.from_llm(
98
+ llm=llm,
99
+ retriever=vectorstore.as_retriever(),
100
+ memory=memory
101
+ )
102
+ return conversation_chain
103
+
104
+
105
+ # ์‚ฌ์šฉ์ž ์ž…๋ ฅ์„ ์ฒ˜๋ฆฌํ•˜๋Š” ํ•จ์ˆ˜์ž…๋‹ˆ๋‹ค.
106
+ def handle_userinput(user_question):
107
+ # ๋Œ€ํ™” ์ฒด์ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ์‚ฌ์šฉ์ž ์งˆ๋ฌธ์— ๋Œ€ํ•œ ์‘๋‹ต์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
108
+ response = st.session_state.conversation({'question': user_question})
109
+ # ๋Œ€ํ™” ๊ธฐ๋ก์„ ์ €์žฅํ•ฉ๋‹ˆ๋‹ค.
110
+ st.session_state.chat_history = response['chat_history']
111
+
112
+ for i, message in enumerate(st.session_state.chat_history):
113
+ if i % 2 == 0:
114
+ st.write(user_template.replace(
115
+ "{{MSG}}", message.content), unsafe_allow_html=True)
116
+ else:
117
+ st.write(bot_template.replace(
118
+ "{{MSG}}", message.content), unsafe_allow_html=True)
119
+
120
+
121
+ def main():
122
+ load_dotenv()
123
+ st.set_page_config(page_title="Chat with multiple Files",
124
+ page_icon=":books:")
125
+ st.write(css, unsafe_allow_html=True)
126
+
127
+ if "conversation" not in st.session_state:
128
+ st.session_state.conversation = None
129
+ if "chat_history" not in st.session_state:
130
+ st.session_state.chat_history = None
131
+
132
+ st.header("Chat with multiple Files :")
133
+ user_question = st.text_input("Ask a question about your documents:")
134
+ if user_question:
135
+ handle_userinput(user_question)
136
+
137
+ with st.sidebar:
138
+ openai_key = st.text_input("Paste your OpenAI API key (sk-...)")
139
+ if openai_key:
140
+ os.environ["OPENAI_API_KEY"] = openai_key
141
+
142
+ st.subheader("Your documents")
143
+ docs = st.file_uploader(
144
+ "Upload your PDFs here and click on 'Process'", accept_multiple_files=True)
145
+ if st.button("Process"):
146
+ with st.spinner("Processing"):
147
+ # get pdf text
148
+ doc_list = []
149
+
150
+ for file in docs:
151
+ print('file - type : ', file.type)
152
+ if file.type == 'text/plain':
153
+ # file is .txt
154
+ doc_list.extend(get_text_file(file))
155
+ elif file.type in ['application/octet-stream', 'application/pdf']:
156
+ # file is .pdf
157
+ doc_list.extend(get_pdf_text(file))
158
+ elif file.type == 'text/csv':
159
+ # file is .csv
160
+ doc_list.extend(get_csv_file(file))
161
+ elif file.type == 'application/json':
162
+ # file is .json
163
+ doc_list.extend(get_json_file(file))
164
+
165
+ # get the text chunks
166
+ text_chunks = get_text_chunks(doc_list)
167
+
168
+ # create vector store
169
+ vectorstore = get_vectorstore(text_chunks)
170
+
171
+ # create conversation chain
172
+ st.session_state.conversation = get_conversation_chain(
173
+ vectorstore)
174
+
175
+
176
+ if __name__ == '__main__':
177
+ main()