urbanManul commited on
Commit
f4552c7
Β·
1 Parent(s): c132146

Upload 3 files

Browse files
Files changed (3) hide show
  1. app.py +177 -0
  2. htmlTemplates.py +44 -0
  3. requirements.txt +13 -0
app.py ADDED
@@ -0,0 +1,177 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+
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
+
23
+ pdf_loader = PyPDFLoader(temp_filepath) # PyPDFLoaderλ₯Ό μ‚¬μš©ν•΄ PDFλ₯Ό λ‘œλ“œν•©λ‹ˆλ‹€.
24
+ pdf_doc = pdf_loader.load() # ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
25
+
26
+ return pdf_doc # μΆ”μΆœν•œ ν…μŠ€νŠΈλ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
27
+
28
+ # 과제
29
+ # μ•„λž˜ ν…μŠ€νŠΈ μΆ”μΆœ ν•¨μˆ˜λ₯Ό μž‘μ„±
30
+ def get_text_file(docs):
31
+ temp_dir = tempfile.TemporaryDirectory()
32
+ temp_filepath = os.path.join(temp_dir.name, docs.name)
33
+
34
+ with open(temp_filepath, 'wb') as f:
35
+ f.write(docs.getvalue())
36
+
37
+ txt_loader = TextLoader(temp_filepath)
38
+ txt_doc = txt_loader.load()
39
+
40
+ return txt_doc
41
+
42
+ def get_csv_file(docs):
43
+ temp_dir = tempfile.TemporaryDirectory()
44
+ temp_filepath = os.path.join(temp_dir.name, docs.name)
45
+
46
+ with open(temp_filepath, 'wb') as f:
47
+ f.write(docs.getvalue())
48
+
49
+ csv_loader = CSVLoader(temp_filepath)
50
+ csv_doc = csv_loader.load()
51
+
52
+ return csv_doc
53
+
54
+ def get_json_file(docs):
55
+ temp_dir = tempfile.TemporaryDirectory()
56
+ temp_filepath = os.path.join(temp_dir.name, docs.name)
57
+
58
+ with open(temp_filepath, 'wb') as f:
59
+ f.write(docs.getvalue())
60
+
61
+ json_loader = JSONLoader(temp_filepath, jq_schema='.clerk[]')
62
+ json_doc = json_loader.load()
63
+
64
+ return json_doc
65
+
66
+
67
+ # λ¬Έμ„œλ“€μ„ μ²˜λ¦¬ν•˜μ—¬ ν…μŠ€νŠΈ 청크둜 λ‚˜λˆ„λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
68
+ def get_text_chunks(documents):
69
+ text_splitter = RecursiveCharacterTextSplitter(
70
+ chunk_size=1000, # 청크의 크기λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
71
+ chunk_overlap=200, # 청크 μ‚¬μ΄μ˜ 쀑볡을 μ§€μ •ν•©λ‹ˆλ‹€.
72
+ length_function=len # ν…μŠ€νŠΈμ˜ 길이λ₯Ό μΈ‘μ •ν•˜λŠ” ν•¨μˆ˜λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
73
+ )
74
+
75
+ documents = text_splitter.split_documents(documents) # λ¬Έμ„œλ“€μ„ 청크둜 λ‚˜λˆ•λ‹ˆλ‹€.
76
+ return documents # λ‚˜λˆˆ 청크λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
77
+
78
+
79
+ # ν…μŠ€νŠΈ μ²­ν¬λ“€λ‘œλΆ€ν„° 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
80
+ def get_vectorstore(text_chunks):
81
+ # μ›ν•˜λŠ” μž„λ² λ”© λͺ¨λΈμ„ λ‘œλ“œν•©λ‹ˆλ‹€.
82
+ embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L12-v2',
83
+ model_kwargs={'device': 'cpu'}) # μž„λ² λ”© λͺ¨λΈμ„ μ„€μ •ν•©λ‹ˆλ‹€.
84
+ vectorstore = FAISS.from_documents(text_chunks, embeddings) # FAISS 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
85
+ return vectorstore # μƒμ„±λœ 벑터 μŠ€ν† μ–΄λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
86
+
87
+
88
+ def get_conversation_chain(vectorstore):
89
+ model_name_or_path = 'TheBloke/Llama-2-7B-chat-GGUF'
90
+ model_basename = 'llama-2-7b-chat.Q2_K.gguf'
91
+ model_path = hf_hub_download(repo_id=model_name_or_path, filename=model_basename)
92
+
93
+ llm = LlamaCpp(model_path=model_path,
94
+ n_ctx=4086,
95
+ input={"temperature": 0.75, "max_length": 2000, "top_p": 1},
96
+ verbose=True, )
97
+ # λŒ€ν™” 기둝을 μ €μž₯ν•˜κΈ° μœ„ν•œ λ©”λͺ¨λ¦¬λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
98
+ memory = ConversationBufferMemory(
99
+ memory_key='chat_history', return_messages=True)
100
+ # λŒ€ν™” 검색 체인을 μƒμ„±ν•©λ‹ˆλ‹€.
101
+ conversation_chain = ConversationalRetrievalChain.from_llm(
102
+ llm=llm,
103
+ retriever=vectorstore.as_retriever(),
104
+ memory=memory
105
+ )
106
+ return conversation_chain # μƒμ„±λœ λŒ€ν™” 체인을 λ°˜ν™˜ν•©λ‹ˆλ‹€.
107
+
108
+ # μ‚¬μš©μž μž…λ ₯을 μ²˜λ¦¬ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
109
+ def handle_userinput(user_question):
110
+ print('user_question => ', user_question)
111
+ # λŒ€ν™” 체인을 μ‚¬μš©ν•˜μ—¬ μ‚¬μš©μž μ§ˆλ¬Έμ— λŒ€ν•œ 응닡을 μƒμ„±ν•©λ‹ˆλ‹€.
112
+ response = st.session_state.conversation({'question': user_question})
113
+ # λŒ€ν™” 기둝을 μ €μž₯ν•©λ‹ˆλ‹€.
114
+ st.session_state.chat_history = response['chat_history']
115
+
116
+ for i, message in enumerate(st.session_state.chat_history):
117
+ if i % 2 == 0:
118
+ st.write(user_template.replace(
119
+ "{{MSG}}", message.content), unsafe_allow_html=True)
120
+ else:
121
+ st.write(bot_template.replace(
122
+ "{{MSG}}", message.content), unsafe_allow_html=True)
123
+
124
+
125
+ def main():
126
+ load_dotenv()
127
+ st.set_page_config(page_title="Chat with multiple Files",
128
+ page_icon=":books:")
129
+ st.write(css, unsafe_allow_html=True)
130
+
131
+ if "conversation" not in st.session_state:
132
+ st.session_state.conversation = None
133
+ if "chat_history" not in st.session_state:
134
+ st.session_state.chat_history = None
135
+
136
+ st.header("Chat with multiple Files:")
137
+ user_question = st.text_input("Ask a question about your documents:")
138
+ if user_question:
139
+ handle_userinput(user_question)
140
+
141
+ with st.sidebar:
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()
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
+ '''
requirements.txt ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ langchain
2
+ llama-cpp-python
3
+ PyPDF2==3.0.1
4
+ faiss-cpu==1.7.4
5
+ ctransformers
6
+ pypdf
7
+ chromadb
8
+ tiktoken
9
+ pysqlite3-binary
10
+ streamlit-extras
11
+ InstructorEmbedding
12
+ sentence-transformers
13
+ jq