LangChain 및 Vertex AI PaLM을 사용한 대규모 문서에 대한 확장 가능한 질문 응답

!pip install google-cloud-aiplatform langchain==0.0.323 chromadb==0.3.26 pypdf

from langchain.document_loaders import PyPDFLoader  
from langchain.llms import VertexAI
from langchain.chains.question_answering import load_qa_chain
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import Chroma
from langchain.embeddings import VertexAIEmbeddings

vertex_llm_text = VertexAI(model_name="text-bison@001")
vertex_embeddings = VertexAIEmbeddings(model_name="textembedding-gecko@001")  

pdf_url = "https://services.google.com/fh/files/misc/practitioners_guide_to_mlops_whitepaper.pdf"
pdf_loader = PyPDFLoader(pdf_file)
pages = pdf_loader.load_and_split()

prompt_template = """제공된 컨텍스트를 사용하여 질문에 가능한 정확하게 답하십시오. 
만약 답변이 컨텍스트에 포함되어 있지 않다면, "컨텍스트에서 답변을 찾을 수 없음"이라고 말하십시오. \n\n
컨텍스트: \n {context}?\n
질문: \n {question} \n
답변:
"""
 
prompt = PromptTemplate(
    template=prompt_template, input_variables=["context", "question"]
)

stuff_chain = load_qa_chain(vertex_llm_text, chain_type="stuff", prompt=prompt)

question = "실험은 무엇인가요?"
context = "\n".join(str(p.page_content) for p in pages[:7])
stuff_answer = stuff_chain(
    {"input_documents": pages[7:10], "question": question}, return_only_outputs=True
)

try:
    print(stuff_chain(
        {"input_documents": pages[7:], "question": question}, 
        return_only_outputs=True))
except Exception as e:  
    print("이 코드는 너무 큰 컨텍스트에서 추론을 실행할 수 없으므로 실패했습니다.")

question_prompt_template = """
제공된 컨텍스트를 사용하여 질문에 가능한 정확하게 답하십시오. \n\n
컨텍스트: \n {context} \n
질문: \n {question} \n  
답변:
"""
question_prompt = PromptTemplate(
    template=question_prompt_template, input_variables=["context", "question"]
)
 
combine_prompt_template = """추출된 콘텐츠와 질문을 사용하여 최종 답변을 만듭니다.  
만약 답변이 컨텍스트에 포함되어 있지 않다면, "컨텍스트에서 답변을 찾을 수 없음"이라고 말하십시오. \n\n
요약: \n {summaries}?\n
질문: \n {question} \n
답변:  
"""
combine_prompt = PromptTemplate(
    template=combine_prompt_template, input_variables=["summaries", "question"]
)

map_reduce_chain = load_qa_chain(
    vertex_llm_text, 
    chain_type="map_reduce",
    return_intermediate_steps=True,
    question_prompt=question_prompt,
    combine_prompt=combine_prompt,
)

map_reduce_outputs = map_reduce_chain({"input_documents": pages, "question": question})

for doc, out in zip(
    map_reduce_outputs["input_documents"], map_reduce_outputs["intermediate_steps"]
):
    print(f"페이지: {doc.metadata['page']}")
    print(f"답변: {out}")

refine_prompt_template = """
The original question is: \n {question} \n
The provided answer is: \n {existing_answer}\n  
Refine the existing answer if needed with the following context: \n {context_str} \n
Given the extracted content and the question, create a final answer.
If the answer is not contained in the context, say "answer not available in context. \n\n  
"""
refine_prompt = PromptTemplate(
    input_variables=["question", "existing_answer", "context_str"],  
    template=refine_prompt_template,
)

refine_chain = load_qa_chain(
    vertex_llm_text,
    chain_type="refine", 
    return_intermediate_steps=True,
    question_prompt=initial_question_prompt,
    refine_prompt=refine_prompt,
)

refine_outputs = refine_chain({"input_documents": pages, "question": question})

for doc, out in zip(
    refine_outputs["input_documents"], refine_outputs["intermediate_steps"]
):
    print(f"페이지: {doc.metadata['page']}")  
    print(f"답변: {out}")

vector_index = Chroma.from_documents(pages, vertex_embeddings).as_retriever()

docs = vector_index.get_relevant_documents(question)

map_reduce_embeddings_outputs = map_reduce_chain(
    {"input_documents": docs, "question": question}
)
print(map_reduce_embeddings_outputs["output_text"])