nijoow commited on
Commit
1b90af4
Β·
1 Parent(s): e76bd22

Upload import streamlit as st.py

Browse files
Files changed (1) hide show
  1. import streamlit as st.py +175 -0
import streamlit as st.py ADDED
@@ -0,0 +1,175 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import streamlit as st
2
+ from dotenv import load_dotenv
3
+ from langchain.text_splitter import CharacterTextSplitter, RecursiveCharacterTextSplitter
4
+ from langchain.vectorstores import FAISS
5
+ from langchain.embeddings import HuggingFaceEmbeddings # General embeddings from HuggingFace models.
6
+ from langchain.memory import ConversationBufferMemory
7
+ from langchain.chains import ConversationalRetrievalChain
8
+ from htmlTemplates import css, bot_template, user_template
9
+ from langchain.llms import LlamaCpp # For loading transformer models.
10
+ from langchain.document_loaders import PyPDFLoader, TextLoader, JSONLoader, CSVLoader
11
+ import tempfile # μž„μ‹œ νŒŒμΌμ„ μƒμ„±ν•˜κΈ° μœ„ν•œ λΌμ΄λΈŒλŸ¬λ¦¬μž…λ‹ˆλ‹€.
12
+ import os
13
+ from huggingface_hub import hf_hub_download # Hugging Face Hubμ—μ„œ λͺ¨λΈμ„ λ‹€μš΄λ‘œλ“œν•˜κΈ° μœ„ν•œ ν•¨μˆ˜μž…λ‹ˆλ‹€.
14
+
15
+
16
+ # PDF λ¬Έμ„œλ‘œλΆ€ν„° ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
17
+ def get_pdf_text(pdf_docs):
18
+ temp_dir = tempfile.TemporaryDirectory() # μž„μ‹œ 디렉토리λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
19
+ temp_filepath = os.path.join(temp_dir.name, pdf_docs.name) # μž„μ‹œ 파일 경둜λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
20
+ with open(temp_filepath, "wb") as f: # μž„μ‹œ νŒŒμΌμ„ λ°”μ΄λ„ˆλ¦¬ μ“°κΈ° λͺ¨λ“œλ‘œ μ—½λ‹ˆλ‹€.
21
+ f.write(pdf_docs.getvalue()) # PDF λ¬Έμ„œμ˜ λ‚΄μš©μ„ μž„μ‹œ νŒŒμΌμ— μ”λ‹ˆλ‹€.
22
+ pdf_loader = PyPDFLoader(temp_filepath) # PyPDFLoaderλ₯Ό μ‚¬μš©ν•΄ PDFλ₯Ό λ‘œλ“œν•©λ‹ˆλ‹€.
23
+ pdf_doc = pdf_loader.load() # ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
24
+ return pdf_doc # μΆ”μΆœν•œ ν…μŠ€νŠΈλ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
25
+
26
+
27
+ # 과제
28
+ # μ•„λž˜ ν…μŠ€νŠΈ μΆ”μΆœ ν•¨μˆ˜λ₯Ό μž‘μ„±
29
+ def get_text_file(docs):
30
+ with NamedTemporaryFile() as temp_file:
31
+ temp_file.write(docs.getvalue())
32
+ temp_file.seek(0)
33
+
34
+ # ν…μŠ€νŠΈ νŒŒμΌμ—μ„œ ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•˜λŠ” λ‘œμ§μ„ κ΅¬ν˜„ν•©λ‹ˆλ‹€.
35
+ text_content = temp_file.read().decode('utf-8')
36
+
37
+ return text_content
38
+
39
+
40
+ import csv
41
+ import json
42
+ from tempfile import NamedTemporaryFile
43
+
44
+ def get_csv_file(docs):
45
+ with NamedTemporaryFile() as temp_file:
46
+ temp_file.write(docs.getvalue())
47
+ temp_file.seek(0)
48
+
49
+ csv_data = []
50
+ csv_reader = csv.reader(temp_file)
51
+ for row in csv_reader:
52
+ csv_data.append(row)
53
+ return csv_data
54
+
55
+ def get_json_file(docs):
56
+ with NamedTemporaryFile() as temp_file:
57
+ temp_file.write(docs.getvalue())
58
+ temp_file.seek(0)
59
+
60
+ json_data = json.load(temp_file)
61
+ return json_data
62
+
63
+
64
+ # λ¬Έμ„œλ“€μ„ μ²˜λ¦¬ν•˜μ—¬ ν…μŠ€νŠΈ 청크둜 λ‚˜λˆ„λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
65
+ def get_text_chunks(documents):
66
+ text_splitter = RecursiveCharacterTextSplitter(
67
+ chunk_size=1000, # 청크의 크기λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
68
+ chunk_overlap=200, # 청크 μ‚¬μ΄μ˜ 쀑볡을 μ§€μ •ν•©λ‹ˆλ‹€.
69
+ length_function=len # ν…μŠ€νŠΈμ˜ 길이λ₯Ό μΈ‘μ •ν•˜λŠ” ν•¨μˆ˜λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
70
+ )
71
+
72
+ documents = text_splitter.split_documents(documents) # λ¬Έμ„œλ“€μ„ 청크둜 λ‚˜λˆ•λ‹ˆλ‹€.
73
+ return documents # λ‚˜λˆˆ 청크λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
74
+
75
+
76
+ # ν…μŠ€νŠΈ μ²­ν¬λ“€λ‘œλΆ€ν„° 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
77
+ def get_vectorstore(text_chunks):
78
+ # μ›ν•˜λŠ” μž„λ² λ”© λͺ¨λΈμ„ λ‘œλ“œν•©λ‹ˆλ‹€.
79
+ embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L12-v2',
80
+ model_kwargs={'device': 'cpu'}) # μž„λ² λ”© λͺ¨λΈμ„ μ„€μ •ν•©λ‹ˆλ‹€.
81
+ vectorstore = FAISS.from_documents(text_chunks, embeddings) # FAISS 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
82
+ return vectorstore # μƒμ„±λœ 벑터 μŠ€ν† μ–΄λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
83
+
84
+
85
+ def get_conversation_chain(vectorstore):
86
+ model_name_or_path = 'TheBloke/Llama-2-7B-chat-GGUF'
87
+ model_basename = 'llama-2-7b-chat.Q2_K.gguf'
88
+ model_path = hf_hub_download(repo_id=model_name_or_path, filename=model_basename)
89
+
90
+ llm = LlamaCpp(model_path=model_path,
91
+ n_ctx=4086,
92
+ input={"temperature": 0.75, "max_length": 2000, "top_p": 1},
93
+ verbose=True, )
94
+ # λŒ€ν™” 기둝을 μ €μž₯ν•˜κΈ° μœ„ν•œ λ©”λͺ¨λ¦¬λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
95
+ memory = ConversationBufferMemory(
96
+ memory_key='chat_history', return_messages=True)
97
+ # λŒ€ν™” 검색 체인을 μƒμ„±ν•©λ‹ˆλ‹€.
98
+ conversation_chain = ConversationalRetrievalChain.from_llm(
99
+ llm=llm,
100
+ retriever=vectorstore.as_retriever(),
101
+ memory=memory
102
+ )
103
+ return conversation_chain # μƒμ„±λœ λŒ€ν™” 체인을 λ°˜ν™˜ν•©λ‹ˆλ‹€.
104
+
105
+
106
+ # μ‚¬μš©μž μž…λ ₯을 μ²˜λ¦¬ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
107
+ def handle_userinput(user_question):
108
+ print('user_question => ', user_question)
109
+ # λŒ€ν™” 체인을 μ‚¬μš©ν•˜μ—¬ μ‚¬μš©μž μ§ˆλ¬Έμ— λŒ€ν•œ 응닡을 μƒμ„±ν•©λ‹ˆλ‹€.
110
+ response = st.session_state.conversation({'question': user_question})
111
+ # λŒ€ν™” 기둝을 μ €μž₯ν•©λ‹ˆλ‹€.
112
+ st.session_state.chat_history = response['chat_history']
113
+
114
+ for i, message in enumerate(st.session_state.chat_history):
115
+ if i % 2 == 0:
116
+ st.write(user_template.replace(
117
+ "{{MSG}}", message.content), unsafe_allow_html=True)
118
+ else:
119
+ st.write(bot_template.replace(
120
+ "{{MSG}}", message.content), unsafe_allow_html=True)
121
+
122
+
123
+ def main():
124
+ load_dotenv()
125
+ st.set_page_config(page_title="Chat with multiple Files",
126
+ page_icon=":books:")
127
+ st.write(css, unsafe_allow_html=True)
128
+
129
+ if "conversation" not in st.session_state:
130
+ st.session_state.conversation = None
131
+ if "chat_history" not in st.session_state:
132
+ st.session_state.chat_history = None
133
+
134
+ st.header("Chat with multiple Files:")
135
+ user_question = st.text_input("Ask a question about your documents:")
136
+ if user_question:
137
+ handle_userinput(user_question)
138
+
139
+ with st.sidebar:
140
+ st.subheader("Your documents")
141
+ docs = st.file_uploader(
142
+ "Upload your PDFs 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
+
174
+ if __name__ == '__main__':
175
+ main()