File size: 4,116 Bytes
1a2426f
 
 
 
 
 
 
 
 
 
 
 
 
 
4888c0f
1a2426f
16c7d1c
1a2426f
 
 
f7a9654
1a2426f
 
 
 
 
a2d3508
1a2426f
 
 
 
 
 
 
 
a2d3508
 
 
1a2426f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d497811
1a2426f
d497811
 
 
1a2426f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4888c0f
1a2426f
 
 
 
 
 
 
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
from langchain.chat_models import ChatOpenAI
from langchain.document_loaders import PyPDFLoader
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.embeddings.cohere import CohereEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores.elastic_vector_search import ElasticVectorSearch
from langchain.vectorstores import Chroma
from PyPDF2 import PdfWriter
import gradio as gr
import os
from dotenv import load_dotenv
import openai

load_dotenv()

os.environ["OPENAI_API_KEY"] = os.environ['my_secret']
llm = ChatOpenAI(model_name="gpt-3.5-turbo-0613", temperature=0)
loader = PyPDFLoader("/home/user/app/docs.pdf")
documents = loader.load()

text_splitter = CharacterTextSplitter(chunk_size=800, chunk_overlap=0)
texts = text_splitter.split_documents(documents)

#vector embedding
embeddings = OpenAIEmbeddings()
vector_store = Chroma.from_documents(texts, embeddings)
retriever = vector_store.as_retriever(search_kwargs={"k": 3})


from langchain.prompts.chat import (
    ChatPromptTemplate,
    SystemMessagePromptTemplate,
    HumanMessagePromptTemplate,
)

system_template="""Your name is Sayno. You act like a successful rich man. 
Never reveal that it is an AI. Use the following pieces of context to answer the users question shortly.
Given the following summaries of a long document and a question, create a final answer with references.
If you don't know the answer, just say that "I don't know", don't try to make up an answer.
----------------
{summaries}

You MUST answer in Korean and in Markdown format:"""

messages = [
    SystemMessagePromptTemplate.from_template(system_template),
    HumanMessagePromptTemplate.from_template("{question}")
]

prompt = ChatPromptTemplate.from_messages(messages)

from langchain.chat_models import ChatOpenAI
from langchain.chains import RetrievalQAWithSourcesChain

chain_type_kwargs = {"prompt": prompt}

chain = RetrievalQAWithSourcesChain.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=retriever,
    return_source_documents=True,
    chain_type_kwargs=chain_type_kwargs,
    reduce_k_below_max_tokens=True,
    verbose=True,
)

query = "ν–‰λ³΅ν•œ μΈμƒμ΄λž€?"
result = chain(query)


for doc in result['source_documents']:
    print('λ‚΄μš© : ' + doc.page_content[0:100].replace('\n', ' '))
    print('파일 : ' + doc.metadata['source'])
    print('νŽ˜μ΄μ§€ : ' + str(doc.metadata['page']))


def respond(message, chat_history):  # μ±„νŒ…λ΄‡μ˜ 응닡을 μ²˜λ¦¬ν•˜λŠ” ν•¨μˆ˜λ₯Ό μ •μ˜ν•©λ‹ˆλ‹€.

    result = chain(message)

    bot_message = result['answer']

    for i, doc in enumerate(result['source_documents']):
        bot_message += '[' + str(i+1) + '] ' + doc.metadata['source'] + '(' + str(doc.metadata['page']) + ') '

    chat_history.append((message, bot_message))  # μ±„νŒ… 기둝에 μ‚¬μš©μžμ˜ λ©”μ‹œμ§€μ™€ λ΄‡μ˜ 응닡을 μΆ”κ°€ν•©λ‹ˆλ‹€.

    return "", chat_history  # μˆ˜μ •λœ μ±„νŒ… 기둝을 λ°˜ν™˜ν•©λ‹ˆλ‹€.

with gr.Blocks(theme='gstaff/sketch') as demo:  # gr.Blocks()λ₯Ό μ‚¬μš©ν•˜μ—¬ μΈν„°νŽ˜μ΄μŠ€λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
    gr.Markdown("# μ•ˆλ…•ν•˜μ„Έμš”. 세이노와 λŒ€ν™”ν•΄λ³΄μ„Έμš”. \n λ‹΅λ³€ 생성에 쑰금 μ‹œκ°„μ΄ μ†Œμš”λ  수 μžˆμŠ΅λ‹ˆλ‹€.")
    chatbot = gr.Chatbot(label="μ±„νŒ…μ°½")  # 'μ±„νŒ…μ°½'μ΄λΌλŠ” λ ˆμ΄λΈ”μ„ 가진 μ±„νŒ…λ΄‡ μ»΄ν¬λ„ŒνŠΈλ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
    msg = gr.Textbox(label="μž…λ ₯")  # 'μž…λ ₯'μ΄λΌλŠ” λ ˆμ΄λΈ”μ„ 가진 ν…μŠ€νŠΈλ°•μŠ€λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
    clear = gr.Button("μ΄ˆκΈ°ν™”")  # 'μ΄ˆκΈ°ν™”'λΌλŠ” λ ˆμ΄λΈ”μ„ 가진 λ²„νŠΌμ„ μƒμ„±ν•©λ‹ˆλ‹€.

    msg.submit(respond, [msg, chatbot], [msg, chatbot])  # ν…μŠ€νŠΈλ°•μŠ€μ— λ©”μ‹œμ§€λ₯Ό μž…λ ₯ν•˜κ³  μ œμΆœν•˜λ©΄ respond ν•¨μˆ˜κ°€ ν˜ΈμΆœλ˜λ„λ‘ ν•©λ‹ˆλ‹€.
    clear.click(lambda: None, None, chatbot, queue=False)  # 'μ΄ˆκΈ°ν™”' λ²„νŠΌμ„ ν΄λ¦­ν•˜λ©΄ μ±„νŒ… 기둝을 μ΄ˆκΈ°ν™”ν•©λ‹ˆλ‹€.
demo.launch(debug=True)  # μΈν„°νŽ˜μ΄μŠ€λ₯Ό μ‹€ν–‰ν•©λ‹ˆλ‹€. μ‹€ν–‰ν•˜λ©΄ μ‚¬μš©μžλŠ” 'μž…λ ₯' ν…μŠ€νŠΈλ°•μŠ€μ— λ©”μ‹œμ§€λ₯Ό μž‘μ„±ν•˜κ³  μ œμΆœν•  수 있으며, 'μ΄ˆκΈ°ν™”' λ²„νŠΌμ„ 톡해 μ±„νŒ… 기둝을 μ΄ˆκΈ°ν™” ν•  수 μžˆμŠ΅λ‹ˆλ‹€.