TypeError: expected string or buffer - Langchain, OpenAI Embeddings

题意:类型错误:期望字符串或缓冲区 - Langchain,OpenAI Embeddings

问题背景:

I am trying to create RAG using the product manuals in pdf which are splitted, indexed and stored in Chroma persisted on a disk. When I try the function that classifies the reviews using the documents context, below is the error I get:

我正在尝试使用 PDF 格式的产品手册创建 RAG,这些手册被拆分、索引并存储在硬盘上的 Chroma 中。当我尝试使用文档上下文对评论进行分类的函数时,出现了以下错误:

复制代码
from langchain import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
from langchain.embeddings import AzureOpenAIEmbeddings
from langchain.chat_models import AzureChatOpenAI
from langchain.vectorstores import Chroma

llm = AzureChatOpenAI(
        azure_deployment="ChatGPT-16K",
        openai_api_version="2023-05-15",
        azure_endpoint=endpoint,
        api_key=result["access_token"],
        temperature=0,
        seed = 100
    )

embedding_model = AzureOpenAIEmbeddings(
    api_version="2023-05-15",
    azure_endpoint=endpoint,
    api_key=result["access_token"],
    azure_deployment="ada002",
)

vectordb = Chroma(
    persist_directory=vector_db_path,
    embedding_function=embedding_model,
    collection_name="product_manuals",
)


def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)

def classify (review_title, review_text, product_num):

    template = """
        
    You are a customer service AI Assistant that handles responses to negative product reviews. 

       Use the context below and categorize {review_title} and {review_text} into defect, misuse or poor quality categories based only on provided context. If you don't know, say that you do not know, don't try to make up an answer. Respond back with an answer in the following format:

        poor quality
        misuse
        defect

        {context}
            
    Category: 
    """


    rag_prompt = PromptTemplate.from_template(template)
    
    retriever = vectordb.as_retriever(search_type="similarity", search_kwargs={'filter': {'product_num': product_num}})


    retrieval_chain = (
            {"context": retriever | format_docs, "review_title: RunnablePassthrough(), "review_text": RunnablePassthrough()}
            | rag_prompt
            | llm
            | StrOutputParser()
    )
    return retrieval_chain.invoke({"review_title": review_title, "review_text": review_text})

classify(review_title="Terrible", review_text ="This baking sheet is terrible. It stains so easily and i've tried everything to get it clean", product_num ="8888999")

Error stack: 错误信息:

复制代码
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
File <command-3066972537097411>, line 1
----> 1 issue_recommendation(
      2     review_title="Terrible",
      3     review_text="This baking sheet is terrible. It stains so easily and i've tried everything to get it clean. I've maybe used it 5 times and it looks like it's 20 years old. The side of the pan also hold water, so when you pick it up off the drying rack, water runs out. I would never purchase these again.",
      4     product_num="8888999"
      5    
      6 )

File <command-3066972537097410>, line 44, in issue_recommendation(review_title, review_text, product_num)
     36 retriever = vectordb.as_retriever(search_type="similarity", search_kwargs={'filter': {'product_num': product_num}})
     38 retrieval_chain = (
     39         {"context": retriever | format_docs, "review_text": RunnablePassthrough()}
     40         | rag_prompt
     41         | llm
     42         | StrOutputParser()
     43 )
---> 44 return retrieval_chain.invoke({"review_title":review_title, "review_text": review_text})

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_core/runnables/base.py:1762, in RunnableSequence.invoke(self, input, config)
   1760 try:
   1761     for i, step in enumerate(self.steps):
-> 1762         input = step.invoke(
   1763             input,
   1764             # mark each step as a child run
   1765             patch_config(
   1766                 config, callbacks=run_manager.get_child(f"seq:step:{i+1}")
   1767             ),
   1768         )
   1769 # finish the root run
   1770 except BaseException as e:

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_core/runnables/base.py:2327, in RunnableParallel.invoke(self, input, config)
   2314     with get_executor_for_config(config) as executor:
   2315         futures = [
   2316             executor.submit(
   2317                 step.invoke,
   (...)
   2325             for key, step in steps.items()
   2326         ]
-> 2327         output = {key: future.result() for key, future in zip(steps, futures)}
   2328 # finish the root run
   2329 except BaseException as e:

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_core/runnables/base.py:2327, in <dictcomp>(.0)
   2314     with get_executor_for_config(config) as executor:
   2315         futures = [
   2316             executor.submit(
   2317                 step.invoke,
   (...)
   2325             for key, step in steps.items()
   2326         ]
-> 2327         output = {key: future.result() for key, future in zip(steps, futures)}
   2328 # finish the root run
   2329 except BaseException as e:

File /usr/lib/python3.10/concurrent/futures/_base.py:451, in Future.result(self, timeout)
    449     raise CancelledError()
    450 elif self._state == FINISHED:
--> 451     return self.__get_result()
    453 self._condition.wait(timeout)
    455 if self._state in [CANCELLED, CANCELLED_AND_NOTIFIED]:

File /usr/lib/python3.10/concurrent/futures/_base.py:403, in Future.__get_result(self)
    401 if self._exception:
    402     try:
--> 403         raise self._exception
    404     finally:
    405         # Break a reference cycle with the exception in self._exception
    406         self = None

File /usr/lib/python3.10/concurrent/futures/thread.py:58, in _WorkItem.run(self)
     55     return
     57 try:
---> 58     result = self.fn(*self.args, **self.kwargs)
     59 except BaseException as exc:
     60     self.future.set_exception(exc)

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_core/runnables/base.py:1762, in RunnableSequence.invoke(self, input, config)
   1760 try:
   1761     for i, step in enumerate(self.steps):
-> 1762         input = step.invoke(
   1763             input,
   1764             # mark each step as a child run
   1765             patch_config(
   1766                 config, callbacks=run_manager.get_child(f"seq:step:{i+1}")
   1767             ),
   1768         )
   1769 # finish the root run
   1770 except BaseException as e:

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_core/retrievers.py:121, in BaseRetriever.invoke(self, input, config)
    117 def invoke(
    118     self, input: str, config: Optional[RunnableConfig] = None
    119 ) -> List[Document]:
    120     config = ensure_config(config)
--> 121     return self.get_relevant_documents(
    122         input,
    123         callbacks=config.get("callbacks"),
    124         tags=config.get("tags"),
    125         metadata=config.get("metadata"),
    126         run_name=config.get("run_name"),
    127     )

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_core/retrievers.py:223, in BaseRetriever.get_relevant_documents(self, query, callbacks, tags, metadata, run_name, **kwargs)
    221 except Exception as e:
    222     run_manager.on_retriever_error(e)
--> 223     raise e
    224 else:
    225     run_manager.on_retriever_end(
    226         result,
    227         **kwargs,
    228     )

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_core/retrievers.py:216, in BaseRetriever.get_relevant_documents(self, query, callbacks, tags, metadata, run_name, **kwargs)
    214 _kwargs = kwargs if self._expects_other_args else {}
    215 if self._new_arg_supported:
--> 216     result = self._get_relevant_documents(
    217         query, run_manager=run_manager, **_kwargs
    218     )
    219 else:
    220     result = self._get_relevant_documents(query, **_kwargs)

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_core/vectorstores.py:654, in VectorStoreRetriever._get_relevant_documents(self, query, run_manager)
    650 def _get_relevant_documents(
    651     self, query: str, *, run_manager: CallbackManagerForRetrieverRun
    652 ) -> List[Document]:
    653     if self.search_type == "similarity":
--> 654         docs = self.vectorstore.similarity_search(query, **self.search_kwargs)
    655     elif self.search_type == "similarity_score_threshold":
    656         docs_and_similarities = (
    657             self.vectorstore.similarity_search_with_relevance_scores(
    658                 query, **self.search_kwargs
    659             )
    660         )

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_community/vectorstores/chroma.py:348, in Chroma.similarity_search(self, query, k, filter, **kwargs)
    331 def similarity_search(
    332     self,
    333     query: str,
   (...)
    336     **kwargs: Any,
    337 ) -> List[Document]:
    338     """Run similarity search with Chroma.
    339 
    340     Args:
   (...)
    346         List[Document]: List of documents most similar to the query text.
    347     """
--> 348     docs_and_scores = self.similarity_search_with_score(
    349         query, k, filter=filter, **kwargs
    350     )
    351     return [doc for doc, _ in docs_and_scores]

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_community/vectorstores/chroma.py:437, in Chroma.similarity_search_with_score(self, query, k, filter, where_document, **kwargs)
    429     results = self.__query_collection(
    430         query_texts=[query],
    431         n_results=k,
   (...)
    434         **kwargs,
    435     )
    436 else:
--> 437     query_embedding = self._embedding_function.embed_query(query)
    438     results = self.__query_collection(
    439         query_embeddings=[query_embedding],
    440         n_results=k,
   (...)
    443         **kwargs,
    444     )
    446 return _results_to_docs_and_scores(results)

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_community/embeddings/openai.py:691, in OpenAIEmbeddings.embed_query(self, text)
    682 def embed_query(self, text: str) -> List[float]:
    683     """Call out to OpenAI's embedding endpoint for embedding query text.
    684 
    685     Args:
   (...)
    689         Embedding for the text.
    690     """
--> 691     return self.embed_documents([text])[0]

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_community/embeddings/openai.py:662, in OpenAIEmbeddings.embed_documents(self, texts, chunk_size)
    659 # NOTE: to keep things simple, we assume the list may contain texts longer
    660 #       than the maximum context and use length-safe embedding function.
    661 engine = cast(str, self.deployment)
--> 662 return self._get_len_safe_embeddings(texts, engine=engine)

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-65a09d8c-062d-4f4f-9c52-1bf534f6511e/lib/python3.10/site-packages/langchain_community/embeddings/openai.py:465, in OpenAIEmbeddings._get_len_safe_embeddings(self, texts, engine, chunk_size)
    459 if self.model.endswith("001"):
    460     # See: https://github.com/openai/openai-python/
    461     #      issues/418#issuecomment-1525939500
    462     # replace newlines, which can negatively affect performance.
    463     text = text.replace("\n", " ")
--> 465 token = encoding.encode(
    466     text=text,
    467     allowed_special=self.allowed_special,
    468     disallowed_special=self.disallowed_special,
    469 )
    471 # Split tokens into chunks respecting the embedding_ctx_length
    472 for j in range(0, len(token), self.embedding_ctx_length):

File /databricks/python/lib/python3.10/site-packages/tiktoken/core.py:116, in Encoding.encode(self, text, allowed_special, disallowed_special)
    114     if not isinstance(disallowed_special, frozenset):
    115         disallowed_special = frozenset(disallowed_special)
--> 116     if match := _special_token_regex(disallowed_special).search(text):
    117         raise_disallowed_special_token(match.group())
    119 try:

TypeError: expected string or buffer

Embeddings seems to work fine when I test. It also works fine when I remove the context and retriever from the chain. It seems to be related to embeddings. Examples on Langchain website instantiates retriver from Chroma.from_documents() whereas I load Chroma vector store from a persisted path. I also tried invoking with review_text only (instead of review title and review text) but the error persists. Not sure why this is happening. These are the package versions I work:

当我测试时,Embeddings 似乎工作正常。当我从链中移除上下文和检索器时,它也能正常工作。问题似乎与 Embeddings 有关。Langchain 网站上的示例是通过 Chroma.from_documents() 实例化检索器,而我是从已保存的路径加载 Chroma 向量存储。我也尝试仅使用 review_text(而不是 review title 和 review text),但错误仍然存在。不确定为什么会这样。这是我使用的包版本:

Name: openai Version: 1.6.1

Name: langchain Version: 0.0.354

问题解决:

I've come across the same issue, and turned out that langchain pass a key-value pair as an input to the encoding.code() while it requires str type. A work around is by using itemgetter() to get the direct string input. It might be something like this

我也遇到了同样的问题,发现是由于 langchain 将一个键值对作为输入传递给 encoding.code(),而它需要的是 str 类型。一个解决方法是使用 itemgetter() 来获取直接的字符串输入。可能是这样的:

复制代码
        retrieval_chain = (
            {
                "document": itemgetter("question") | self.retriever,
                "question": itemgetter("question"),
            }
            | prompt
            | model
            | StrOutputParser()
        )

You can find the reference here

你可以在这里找到参考资料。

相关推荐
ComPDFKit16 小时前
RAG 回答只显示 PDF 文件名?用 Python 保留引用页码
开发语言·python·pdf·rag
YIAN18 小时前
从 LangChain 到 DeepAgents:手把手拆解 createAgent、Middleware 与复杂 Agent
javascript·langchain·deepseek
K72 天前
别再搜「LangChain 教程」了——你学的 Chain 已经被判了死刑
langchain·ai编程
xyphf_和派孔明2 天前
第六章: AI Agent长期记忆的存储引擎介绍
langchain·llm·agent
xyphf_和派孔明2 天前
第四章: AI Agent短期记忆的信息管理机制设计
langchain·llm·agent
Soari2 天前
NetPilot:用RAG与验证回路辅助混合TSN和5G网络配置
5g·网络配置·大语言模型·tsn·rag
叼烟扛炮2 天前
[LangChain] 学习笔记(13):并发不是越大越好——batch / gather / 信号量 / 配额治理
笔记·学习·langchain
攒了一袋星辰2 天前
AI 求职长期记忆原理解析:从碎片经历到贝叶斯能力星图
知识图谱·agent·rag·ai工程·校招面试
叼烟扛炮3 天前
[LangChain] 学习笔记(09):工具(Tool)定义与调用闭环 · 理论+实战完整版
笔记·学习·langchain
LiaoWL1233 天前
从零搭一个 RAG 知识库(二):换上语义 Embedding——BGE 实测与阈值重校准
python·大模型·numpy·embedding·向量检索·rag