File size: 7,290 Bytes
a2f65b2
 
 
 
 
 
 
 
 
 
 
 
 
e770114
1d43ab0
ba2ba98
a2f65b2
 
 
 
 
 
 
 
 
 
 
 
 
10182d9
4505847
8d921b1
ef30a5d
ba2ba98
ef30a5d
 
bd29c3c
64a2b6b
ba2ba98
a2f65b2
ba2ba98
 
 
 
 
 
a2f65b2
 
ba2ba98
 
 
 
 
 
 
 
 
 
 
 
 
a2f65b2
b035b13
 
 
 
 
 
a2f65b2
d32b31c
b035b13
a2f65b2
d32b31c
a2f65b2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
import streamlit as st
from dotenv import load_dotenv
from langchain.text_splitter import CharacterTextSplitter, RecursiveCharacterTextSplitter
from langchain.vectorstores import FAISS
from langchain.embeddings import HuggingFaceEmbeddings  # General embeddings from HuggingFace models.
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain
from htmlTemplates import css, bot_template, user_template
from langchain.llms import LlamaCpp  # For loading transformer models.
from langchain.document_loaders import PyPDFLoader, TextLoader, JSONLoader, CSVLoader
import tempfile # μž„μ‹œ νŒŒμΌμ„ μƒμ„±ν•˜κΈ° μœ„ν•œ λΌμ΄λΈŒλŸ¬λ¦¬μž…λ‹ˆλ‹€.
import os
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‹€.
from transformers import pipeline
from io import BytesIO
import pandas as pd

# PDF λ¬Έμ„œλ‘œλΆ€ν„° ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def get_pdf_text(pdf_docs):
    temp_dir = tempfile.TemporaryDirectory() # μž„μ‹œ 디렉토리λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
    temp_filepath = os.path.join(temp_dir.name, pdf_docs.name) # μž„μ‹œ 파일 경둜λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
    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‹€.
        f.write(pdf_docs.getvalue()) # PDF λ¬Έμ„œμ˜ λ‚΄μš©μ„ μž„μ‹œ νŒŒμΌμ— μ”λ‹ˆλ‹€.
    pdf_loader = PyPDFLoader(temp_filepath) # PyPDFLoaderλ₯Ό μ‚¬μš©ν•΄ PDFλ₯Ό λ‘œλ“œν•©λ‹ˆλ‹€.
    pdf_doc = pdf_loader.load() # ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
    return pdf_doc # μΆ”μΆœν•œ ν…μŠ€νŠΈλ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.

# 과제
# μ•„λž˜ ν…μŠ€νŠΈ μΆ”μΆœ ν•¨μˆ˜λ₯Ό μž‘μ„±
# ν…μŠ€νŠΈ νŒŒμΌλ‘œλΆ€ν„° ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def get_text_file(docs):
    text_list = []
    for doc in docs:
        text = doc.getvalue().decode('utf-8')  # 'BytesIO' κ°μ²΄μ—μ„œ ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
        text_list.append(text)
    return text_list



def get_csv_file(docs):
    text_list = []
    for doc in docs:
        df = pd.read_csv(doc)
        for column in df.columns:
            text_list.extend(df[column].astype(str).tolist())
    return text_list

def get_json_file(docs):
    text_list = []
    for doc in docs:
        json_data = doc.read().decode('utf-8')
        data = json.loads(json_data)
        if isinstance(data, dict):
            text_list.extend(list(data.values()))
        elif isinstance(data, list):
            for item in data:
                if isinstance(item, str):
                    text_list.append(item)
                elif isinstance(item, dict):
                    text_list.extend(list(item.values()))
    return text_list

def get_text_chunks(documents):
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=1000,  # 청크의 크기λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
        chunk_overlap=200,  # 청크 μ‚¬μ΄μ˜ 쀑볡을 μ§€μ •ν•©λ‹ˆλ‹€.
        length_function=len  # ν…μŠ€νŠΈμ˜ 길이λ₯Ό μΈ‘μ •ν•˜λŠ” ν•¨μˆ˜λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
    )

    documents = text_splitter.split_documents(documents)  # λ¬Έμ„œλ“€μ„ 청크둜 λ‚˜λˆ•λ‹ˆλ‹€.
    return documents  # λ‚˜λˆˆ 청크λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.


# ν…μŠ€νŠΈ μ²­ν¬λ“€λ‘œλΆ€ν„° 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def get_vectorstore(text_chunks):
    # μ›ν•˜λŠ” μž„λ² λ”© λͺ¨λΈμ„ λ‘œλ“œν•©λ‹ˆλ‹€.
    embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L12-v2',
                                       model_kwargs={'device': 'cpu'})  # μž„λ² λ”© λͺ¨λΈμ„ μ„€μ •ν•©λ‹ˆλ‹€.
    vectorstore = FAISS.from_documents(text_chunks, embeddings)  # FAISS 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
    return vectorstore  # μƒμ„±λœ 벑터 μŠ€ν† μ–΄λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.


def get_conversation_chain(vectorstore):
    model_name_or_path = 'TheBloke/Llama-2-7B-chat-GGUF'
    model_basename = 'llama-2-7b-chat.Q2_K.gguf'
    model_path = hf_hub_download(repo_id=model_name_or_path, filename=model_basename)

    llm = LlamaCpp(model_path=model_path,
                   n_ctx=4086,
                   input={"temperature": 0.75, "max_length": 2000, "top_p": 1},
                   verbose=True, )
    # λŒ€ν™” 기둝을 μ €μž₯ν•˜κΈ° μœ„ν•œ λ©”λͺ¨λ¦¬λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
    memory = ConversationBufferMemory(
        memory_key='chat_history', return_messages=True)
    # λŒ€ν™” 검색 체인을 μƒμ„±ν•©λ‹ˆλ‹€.
    conversation_chain = ConversationalRetrievalChain.from_llm(
        llm=llm,
        retriever=vectorstore.as_retriever(),
        memory=memory
    )
    return conversation_chain # μƒμ„±λœ λŒ€ν™” 체인을 λ°˜ν™˜ν•©λ‹ˆλ‹€.

# μ‚¬μš©μž μž…λ ₯을 μ²˜λ¦¬ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def handle_userinput(user_question):
    print('user_question =>  ', user_question)
    # λŒ€ν™” 체인을 μ‚¬μš©ν•˜μ—¬ μ‚¬μš©μž μ§ˆλ¬Έμ— λŒ€ν•œ 응닡을 μƒμ„±ν•©λ‹ˆλ‹€.
    response = st.session_state.conversation({'question': user_question})
    # λŒ€ν™” 기둝을 μ €μž₯ν•©λ‹ˆλ‹€.
    st.session_state.chat_history = response['chat_history']

    for i, message in enumerate(st.session_state.chat_history):
        if i % 2 == 0:
            st.write(user_template.replace(
                "{{MSG}}", message.content), unsafe_allow_html=True)
        else:
            st.write(bot_template.replace(
                "{{MSG}}", message.content), unsafe_allow_html=True)


def main():
    load_dotenv()
    st.set_page_config(page_title="Chat with multiple Files",
                       page_icon=":books:")
    st.write(css, unsafe_allow_html=True)

    if "conversation" not in st.session_state:
        st.session_state.conversation = None
    if "chat_history" not in st.session_state:
        st.session_state.chat_history = None

    st.header("Chat with multiple Files:")
    user_question = st.text_input("Ask a question about your documents:")
    if user_question:
        handle_userinput(user_question)

    with st.sidebar:
        st.subheader("Your documents")
        docs = st.file_uploader(
            "Upload your PDFs here and click on 'Process'", accept_multiple_files=True)
        if st.button("Process"):
            with st.spinner("Processing"):
                # get pdf text
                doc_list = []

                for file in docs:
                    print('file - type : ', file.type)
                    if file.type == 'text/plain':
                        # file is .txt
                        doc_list.extend(get_text_file(file))
                    elif file.type in ['application/octet-stream', 'application/pdf']:
                        # file is .pdf
                        doc_list.extend(get_pdf_text(file))
                    elif file.type == 'text/csv':
                        # file is .csv
                        doc_list.extend(get_csv_file(file))
                    elif file.type == 'application/json':
                        # file is .json
                        doc_list.extend(get_json_file(file))

                # get the text chunks
                text_chunks = get_text_chunks(doc_list)

                # create vector store
                vectorstore = get_vectorstore(text_chunks)

                # create conversation chain
                st.session_state.conversation = get_conversation_chain(
                    vectorstore)


if __name__ == '__main__':
    main()