DEVOCEAN

$ ollama list
NAME             	ID          	SIZE  	MODIFIED
ggml-q5-km:latest	c3ebb32b93a5	7.7 GB	2 weeks ago
$

import os
from langchain.document_loaders import TextLoader

import warnings
warnings.filterwarnings("ignore")

############################################################################
## TextLoader
############################################################################
#loader = TextLoader('little.prince.txt')
#pages = loader.load()

from langchain_community.document_loaders import PyMuPDFLoader
loader = PyMuPDFLoader("2024.kbo.rules.pdf")
pages = loader.load()

text = pages[0].page_content

import tiktoken
from langchain.text_splitter import RecursiveCharacterTextSplitter


# 문장 임베딩 및 벡터 저장소 생성
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import Chroma

############################################################################
## TextSplitter
############################################################################
# 문서를 문장으로 분리
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=500,
    chunk_overlap=50,
)
docs = text_splitter.split_documents(pages)
#print(docs[0])
#print(docs[1])


############################################################################
## Text Vector Embedding
############################################################################
# 문장을 임베딩으로 변환하고 벡터 저장소에 저장
embeddings = HuggingFaceEmbeddings(
    #model_name='jhgan/ko-sroberta-nli',
    model_name='BAAI/bge-m3',
    #model_kwargs={'device':'cpu'},
    model_kwargs={'device':'mps'},
    encode_kwargs={'normalize_embeddings':True},
)


############################################################################
## VectorStore(Chroma)
############################################################################
vectorstore = Chroma.from_documents(docs, embeddings)


# 벡터 저장소 경로 설정
vectorstore_path = 'vectorstore'
os.makedirs(vectorstore_path, exist_ok=True)

# 벡터 저장소 생성 및 저장
vectorstore = Chroma.from_documents(docs, embeddings, persist_directory=vectorstore_path)
vectorstore.persist()
print("Vectorstore created and persisted")




############################################################################
## retriever
############################################################################
from langchain_core.runnables import RunnablePassthrough
from langchain_community.chat_models import ChatOllama

model = ChatOllama(model="ggml-q5-km", temperature=0)

# 검색 쿼리
query = "승률계산법을 알려 줘"

retriever = vectorstore.as_retriever(search_kwargs={'k': 3})

docs = retriever.get_relevant_documents(query)
#print(len(docs))
#print(docs[0].page_content)
#print(docs[0].metadata)


from langchain_community.chat_models import ChatOllama
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder

# Prompt
template = '''Answer the question in korean based only on the following context:
{context}

Question: {question}
'''

prompt = ChatPromptTemplate.from_template(template)

def format_docs(docs):
    return '\n\n'.join([d.page_content for d in docs])

# RAG Chain 연결
rag_chain = (
    {'context': retriever | format_docs, 'question': RunnablePassthrough()}
    | prompt
    | model
    | StrOutputParser()
)

# Chain 실행
query = "승률계산법을 알려 줘"
answer = rag_chain.invoke(query)

print("Answer:", answer)

$ python rag_concept.py
Answer: 승률을 계산하기 위해서는 다음과 같은 공식을 사용할 수 있습니다:

승률 = (승리수 / 경기수) x 100

여기서 승리수는 팀이 이긴 게임의 수를, 경기수는 총 플레이한 게임의 수를 의미합니다. 예를 들어, 팀이 60경기 중 40경기를 이겼다면, 그 팀의 승률을 계산하려면 다음과 같이 하면 됩니다:

승률 = (40 / 60) x 100 = 67%

이는 해당 팀이 시즌 동안 67%의 경기에서 승리했음을 의미합니다.