GoBoKyung commited on
Commit
d17abbf
β€’
1 Parent(s): e718cfe
Files changed (2) hide show
  1. app.py +157 -0
  2. htmlTemplates.py +44 -0
app.py ADDED
@@ -0,0 +1,157 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ # PDF λ¬Έμ„œλ‘œλΆ€ν„° ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
16
+ def get_pdf_text(pdf_docs):
17
+ temp_dir = tempfile.TemporaryDirectory() # μž„μ‹œ 디렉토리λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
18
+ temp_filepath = os.path.join(temp_dir.name, pdf_docs.name) # μž„μ‹œ 파일 경둜λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
19
+ 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‹€.
20
+ f.write(pdf_docs.getvalue()) # PDF λ¬Έμ„œμ˜ λ‚΄μš©μ„ μž„μ‹œ νŒŒμΌμ— μ”λ‹ˆλ‹€.
21
+ pdf_loader = PyPDFLoader(temp_filepath) # PyPDFLoaderλ₯Ό μ‚¬μš©ν•΄ PDFλ₯Ό λ‘œλ“œν•©λ‹ˆλ‹€.
22
+ pdf_doc = pdf_loader.load() # ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
23
+ return pdf_doc # μΆ”μΆœν•œ ν…μŠ€νŠΈλ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
24
+
25
+ # 과제
26
+ # μ•„λž˜ ν…μŠ€νŠΈ μΆ”μΆœ ν•¨μˆ˜λ₯Ό μž‘μ„±
27
+ def get_text_file(docs):
28
+ text_loader = TextLoader(docs) # TextLoaderλ₯Ό μ‚¬μš©ν•΄ ν…μŠ€νŠΈ νŒŒμΌμ„ λ‘œλ“œν•©λ‹ˆλ‹€.
29
+ text = text_loader.load() # ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
30
+ return text # μΆ”μΆœν•œ ν…μŠ€νŠΈλ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
31
+
32
+ def get_csv_file(docs):
33
+ csv_loader = CSVLoader(docs) # CSVLoaderλ₯Ό μ‚¬μš©ν•΄ CSV νŒŒμΌμ„ λ‘œλ“œν•©λ‹ˆλ‹€.
34
+ csv_data = csv_loader.load() # CSV 데이터λ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
35
+ text_column = csv_data['your_text_column'] # νŠΉμ • ν…μŠ€νŠΈ μ»¬λŸΌμ„ μ„ νƒν•©λ‹ˆλ‹€.
36
+ return text_column.tolist() # ν…μŠ€νŠΈ μ»¬λŸΌμ„ 리슀트둜 λ°˜ν™˜ν•©λ‹ˆλ‹€.
37
+
38
+ def get_json_file(docs):
39
+ json_loader = JSONLoader(docs) # JSONLoaderλ₯Ό μ‚¬μš©ν•΄ JSON νŒŒμΌμ„ λ‘œλ“œν•©λ‹ˆλ‹€.
40
+ json_data = json_loader.load() # JSON 데이터λ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
41
+ text_field = json_data['your_text_field'] # νŠΉμ • ν…μŠ€νŠΈ ν•„λ“œλ₯Ό μ„ νƒν•©λ‹ˆλ‹€.
42
+ if isinstance(text_field, list):
43
+ return text_field # ν…μŠ€νŠΈ ν•„λ“œκ°€ 리슀트인 경우 λ°”λ‘œ λ°˜ν™˜ν•©λ‹ˆλ‹€.
44
+ else:
45
+ return [text_field] # ν…μŠ€νŠΈ ν•„λ“œκ°€ 단일 값인 경우 리슀트둜 λ³€ν™˜ν•˜μ—¬ λ°˜ν™˜ν•©λ‹ˆλ‹€.
46
+
47
+ # λ¬Έμ„œλ“€μ„ μ²˜λ¦¬ν•˜μ—¬ ν…μŠ€νŠΈ 청크둜 λ‚˜λˆ„λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
48
+ def get_text_chunks(documents):
49
+ text_splitter = RecursiveCharacterTextSplitter(
50
+ chunk_size=1000, # 청크의 크기λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
51
+ chunk_overlap=200, # 청크 μ‚¬μ΄μ˜ 쀑볡을 μ§€μ •ν•©λ‹ˆλ‹€.
52
+ length_function=len # ν…μŠ€νŠΈμ˜ 길이λ₯Ό μΈ‘μ •ν•˜λŠ” ν•¨μˆ˜λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
53
+ )
54
+
55
+ documents = text_splitter.split_documents(documents) # λ¬Έμ„œλ“€μ„ 청크둜 λ‚˜λˆ•λ‹ˆλ‹€.
56
+ return documents # λ‚˜λˆˆ 청크λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
57
+
58
+
59
+ # ν…μŠ€νŠΈ μ²­ν¬λ“€λ‘œλΆ€ν„° 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
60
+ def get_vectorstore(text_chunks):
61
+ # μ›ν•˜λŠ” μž„λ² λ”© λͺ¨λΈμ„ λ‘œλ“œν•©λ‹ˆλ‹€.
62
+ embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L12-v2',
63
+ model_kwargs={'device': 'cpu'}) # μž„λ² λ”© λͺ¨λΈμ„ μ„€μ •ν•©λ‹ˆλ‹€.
64
+ vectorstore = FAISS.from_documents(text_chunks, embeddings) # FAISS 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
65
+ return vectorstore # μƒμ„±λœ 벑터 μŠ€ν† μ–΄λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
66
+
67
+
68
+ def get_conversation_chain(vectorstore):
69
+ model_name_or_path = 'TheBloke/Llama-2-7B-chat-GGUF'
70
+ model_basename = 'llama-2-7b-chat.Q2_K.gguf'
71
+ model_path = hf_hub_download(repo_id=model_name_or_path, filename=model_basename)
72
+
73
+ llm = LlamaCpp(model_path=model_path,
74
+ n_ctx=4086,
75
+ input={"temperature": 0.75, "max_length": 2000, "top_p": 1},
76
+ verbose=True, )
77
+ # λŒ€ν™” 기둝을 μ €μž₯ν•˜κΈ° μœ„ν•œ λ©”λͺ¨λ¦¬λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
78
+ memory = ConversationBufferMemory(
79
+ memory_key='chat_history', return_messages=True)
80
+ # λŒ€ν™” 검색 체인을 μƒμ„±ν•©λ‹ˆλ‹€.
81
+ conversation_chain = ConversationalRetrievalChain.from_llm(
82
+ llm=llm,
83
+ retriever=vectorstore.as_retriever(),
84
+ memory=memory
85
+ )
86
+ return conversation_chain # μƒμ„±λœ λŒ€ν™” 체인을 λ°˜ν™˜ν•©λ‹ˆλ‹€.
87
+
88
+ # μ‚¬μš©μž μž…λ ₯을 μ²˜λ¦¬ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
89
+ def handle_userinput(user_question):
90
+ print('user_question => ', user_question)
91
+ # λŒ€ν™” 체인을 μ‚¬μš©ν•˜μ—¬ μ‚¬μš©μž μ§ˆλ¬Έμ— λŒ€ν•œ 응닡을 μƒμ„±ν•©λ‹ˆλ‹€.
92
+ response = st.session_state.conversation({'question': user_question})
93
+ # λŒ€ν™” 기둝을 μ €μž₯ν•©λ‹ˆλ‹€.
94
+ st.session_state.chat_history = response['chat_history']
95
+
96
+ for i, message in enumerate(st.session_state.chat_history):
97
+ if i % 2 == 0:
98
+ st.write(user_template.replace(
99
+ "{{MSG}}", message.content), unsafe_allow_html=True)
100
+ else:
101
+ st.write(bot_template.replace(
102
+ "{{MSG}}", message.content), unsafe_allow_html=True)
103
+
104
+
105
+ def main():
106
+ load_dotenv()
107
+ st.set_page_config(page_title="Chat with multiple Files",
108
+ page_icon=":books:")
109
+ st.write(css, unsafe_allow_html=True)
110
+
111
+ if "conversation" not in st.session_state:
112
+ st.session_state.conversation = None
113
+ if "chat_history" not in st.session_state:
114
+ st.session_state.chat_history = None
115
+
116
+ st.header("Chat with multiple Files:")
117
+ user_question = st.text_input("Ask a question about your documents:")
118
+ if user_question:
119
+ handle_userinput(user_question)
120
+
121
+ with st.sidebar:
122
+ st.subheader("Your documents")
123
+ docs = st.file_uploader(
124
+ "Upload your PDFs here and click on 'Process'", accept_multiple_files=True)
125
+ if st.button("Process"):
126
+ with st.spinner("Processing"):
127
+ # get pdf text
128
+ doc_list = []
129
+
130
+ for file in docs:
131
+ print('file - type : ', file.type)
132
+ if file.type == 'text/plain':
133
+ # file is .txt
134
+ doc_list.extend(get_text_file(file))
135
+ elif file.type in ['application/octet-stream', 'application/pdf']:
136
+ # file is .pdf
137
+ doc_list.extend(get_pdf_text(file))
138
+ elif file.type == 'text/csv':
139
+ # file is .csv
140
+ doc_list.extend(get_csv_file(file))
141
+ elif file.type == 'application/json':
142
+ # file is .json
143
+ doc_list.extend(get_json_file(file))
144
+
145
+ # get the text chunks
146
+ text_chunks = get_text_chunks(doc_list)
147
+
148
+ # create vector store
149
+ vectorstore = get_vectorstore(text_chunks)
150
+
151
+ # create conversation chain
152
+ st.session_state.conversation = get_conversation_chain(
153
+ vectorstore)
154
+
155
+
156
+ if __name__ == '__main__':
157
+ main()
htmlTemplates.py ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ css = '''
2
+ <style>
3
+ .chat-message {
4
+ padding: 1.5rem; border-radius: 0.5rem; margin-bottom: 1rem; display: flex
5
+ }
6
+ .chat-message.user {
7
+ background-color: #2b313e
8
+ }
9
+ .chat-message.bot {
10
+ background-color: #475063
11
+ }
12
+ .chat-message .avatar {
13
+ width: 20%;
14
+ }
15
+ .chat-message .avatar img {
16
+ max-width: 78px;
17
+ max-height: 78px;
18
+ border-radius: 50%;
19
+ object-fit: cover;
20
+ }
21
+ .chat-message .message {
22
+ width: 80%;
23
+ padding: 0 1.5rem;
24
+ color: #fff;
25
+ }
26
+ '''
27
+
28
+ bot_template = '''
29
+ <div class="chat-message bot">
30
+ <div class="avatar">
31
+ <img src="https://i.ibb.co/cN0nmSj/Screenshot-2023-05-28-at-02-37-21.png" style="max-height: 78px; max-width: 78px; border-radius: 50%; object-fit: cover;">
32
+ </div>
33
+ <div class="message">{{MSG}}</div>
34
+ </div>
35
+ '''
36
+
37
+ user_template = '''
38
+ <div class="chat-message user">
39
+ <div class="avatar">
40
+ <img src="https://i.ibb.co/rdZC7LZ/Photo-logo-1.png">
41
+ </div>
42
+ <div class="message">{{MSG}}</div>
43
+ </div>
44
+ '''