评估链判定正确答案和模型答案是否一致

from langchain.chains import RetrievalQA
from langchain.evaluation import QAEvalChain
from langchain_community.document_loaders import TextLoader
from langchain_community.embeddings import OllamaEmbeddings
from langchain_community.llms.ollama import Ollama
from langchain_community.vectorstores.faiss import FAISS    #使用 vectorstore 进行存储
from langchain_text_splitters import RecursiveCharacterTextSplitter

llm = Ollama(model="qwen:7b")

loader = TextLoader("/home/cmcc/server/file/测试文档.txt", encoding="utf-8")

doc = loader.load()

text_splitter = RecursiveCharacterTextSplitter(chunk_size=3000, chunk_overlap=400)

docs = text_splitter.split_documents(doc)

#向量搜索引擎
embeddings = OllamaEmbeddings()

docsearch = FAISS.from_documents(docs, embeddings)
# FAISS.save_local("Ver", "index")

#创建你的检索引擎  retriever 设置检索的数据库是那个
chain = RetrievalQA.from_chain_type(llm = llm, chain_type="stuff", retriever=docsearch.as_retriever(), input_key = "question")

question_answers = [
    {"question":"小明的哥哥是谁", "answer":"小雄"},
    {"question":"小刘的哥哥是谁", "answer":"小蓝"},
    {"question":"小张的姐姐是谁", "answer":"小红"},

]

predictions = chain.apply(question_answers)
print(predictions)
#启动评估链
eval_chain = QAEvalChain.from_llm(llm)

#让模型自己评分，下面代码帮助 eval_chain 指导不同部分在哪里
graed_outputs = eval_chain.evaluate(question_answers,
                                    predictions,
                                    question_key="question",
                                    prediction_key="result",
                                    answer_key="answer"
                                    )
print(graed_outputs)

# query = "小张的姐姐是谁"
# response = chain.run(query)
# print(response)