使用Pinecone构建RetrievalQA时BaseRetriever实例化失败求助

尝试结合本地LLM与Pinecone向量库构建RetrievalQA链时，出现以下验证错误：

ValidationError                           Traceback (most recent call last)
/Users/dhruv/Desktop/Machine_Learning/Projects/Medical_ChatBot_Application/preprocessing/Experiment.ipynb Cell 28 line 1
----> 1 qa=RetrievalQA.from_chain_type(
      2     llm=llm, 
      3     chain_type="stuff", 
      4     retriever=docsearch.as_retriever(search_kwargs={'k': 2}),
      5     return_source_documents=True, 
      6     chain_type_kwargs=chain_type_kwargs)

File ~/anaconda3/envs/mchatbot/lib/python3.9/site-packages/langchain/chains/retrieval_qa/base.py:95, in BaseRetrievalQA.from_chain_type(cls, llm, chain_type, chain_type_kwargs, **kwargs)
     91 _chain_type_kwargs = chain_type_kwargs or {}
     92 combine_documents_chain = load_qa_chain(
     93     llm, chain_type=chain_type, **_chain_type_kwargs
     94 )
---> 95 return cls(combine_documents_chain=combine_documents_chain, **kwargs)

File ~/anaconda3/envs/mchatbot/lib/python3.9/site-packages/langchain/load/serializable.py:74, in Serializable.__init__(self, **kwargs) -> None:
     73 def __init__(self, **kwargs: Any) -> None:
---> 74     super().__init__(**kwargs)
     75     self._lc_kwargs = kwargs

File ~/anaconda3/envs/mchatbot/lib/python3.9/site-packages/pydantic/main.py:341, in pydantic.main.BaseModel.__init__()

ValidationError: 1 validation error for RetrievalQA
retriever
  Can't instantiate abstract class BaseRetriever with abstract methods _aget_relevant_documents, _get_relevant_documents (type=type_error)

当前docsearch实例通过以下方式创建：

from langchain_pinecone import PineconeVectorStore as PC

docsearch = PC.from_texts([t.page_content for t in text_chunks],
        embeddings,
        index_name = index_name)

解决方案 #

1. 升级依赖包版本 #

版本不兼容是常见触发原因，执行以下命令升级到最新稳定版：

pip install --upgrade langchain langchain-pinecone pinecone-client

2. 手动实例化PineconeRetriever #

避免使用as_retriever方法，直接构建检索器实例：

from langchain_pinecone import PineconeRetriever
from langchain.chains import RetrievalQA

# 初始化检索器
retriever = PineconeRetriever(
    index_name=index_name,
    embedding=embeddings,
    search_kwargs={'k': 2}
)

# 创建RetrievalQA链
qa = RetrievalQA.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=retriever,
    return_source_documents=True,
    chain_type_kwargs=chain_type_kwargs
)

3. 验证向量索引维度匹配 #

确保Pinecone索引的维度与你使用的嵌入模型输出维度完全一致，例如使用SentenceTransformer嵌入时通常为768维，需对应创建索引时的维度设置。

内容的提问来源于stack exchange，提问作者Dhruv Pamneja