LCEL-LangChain RAG顺序链连接报错：expected string or buffer

• 使用LCEL-LangChain-Python构建RAG顺序链，单独运行各链功能正常，但连接成整体链后触发expected string or buffer错误
• 已确认所有相关变量为str类型，问题仍未解决

原始代码 #

from source.main.create_db import vectordb 
from langchain.llms import OpenAI
from langchain import prompts
from langchain.prompts import PromptTemplate
from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI
from langchain import memory
from dotenv import load_dotenv
import os as os
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnableParallel, RunnablePassthrough
from langchain_anthropic import ChatAnthropic
from langchain_core.documents import Document


load_dotenv()
key = os.getenv('OPENAI_API_KEY')
llm = OpenAI()
retriever = vectordb.as_retriever()

template = '''
ou are an assistant specialized in Brazilian public exams, 
mainly focused on tests in the field of law,
and must answer the questions below
{question}
'''

prompt= ChatPromptTemplate.from_template(template)
output_parser = StrOutputParser()
model = llm

chain1 = prompt | model | output_parser 


template2 = '''
Correct and enhance the {answer} using your new information from the 
{context} and provide more references and support for the newly crafted response."
'''

prompt2 =  ChatPromptTemplate.from_template(template2)
setup_and_retrieval = RunnableParallel(
    {'context':retriever , "answer":RunnablePassthrough()}
)

chain2 = ({'answer': chain1} | setup_and_retrieval | prompt2 | model | StrOutputParser() )


a = chain2.invoke('tell me one law')

print(a)

报错信息 #

a = chain2.invoke('tell me one law')
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_core/runnables/base.py", line 2499, in invoke
    input = step.invoke(
            ^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_core/runnables/base.py", line 3144, in invoke
    output = {key: future.result() for key, future in zip(steps, futures)}
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_core/runnables/base.py", line 3144, in <dictcomp>
    output = {key: future.result() for key, future in zip(steps, futures)}
                   ^^^^^^^^^^^^^^^
  File "/opt/anaconda3/lib/python3.11/concurrent/futures/_base.py", line 449, in result
    return self.__get_result()
           ^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/lib/python3.11/concurrent/futures/_base.py", line 401, in __get_result
    raise self._exception
  File "/opt/anaconda3/lib/python3.11/concurrent/futures/thread.py", line 58, in run
    result = self.fn(*self.args, **self.kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_core/retrievers.py", line 141, in invoke
    return self.get_relevant_documents(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_core/retrievers.py", line 245, in get_relevant_documents
    raise e
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_core/retrievers.py", line 238, in get_relevant_documents
    result = self._get_relevant_documents(
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_core/vectorstores.py", line 696, in _get_relevant_documents
    docs = self.vectorstore.similarity_search(query, **self.search_kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_community/vectorstores/chroma.py", line 348, in similarity_search
    docs_and_scores = self.similarity_search_with_score(
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_community/vectorstores/chroma.py", line 437, in similarity_search_with_score
    query_embedding = self._embedding_function.embed_query(query)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_openai/embeddings/base.py", line 546, in embed_query
    return self.embed_documents([text])[0]
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_openai/embeddings/base.py", line 517, in embed_documents
    return self._get_len_safe_embeddings(texts, engine=engine)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/langchain_openai/embeddings/base.py", line 310, in _get_len_safe_embeddings
    token = encoding.encode(
            ^^^^^^^^^^^^^^^^
  File "/Users/erickrochaleaodasilva/projeto_de_lei/maquina/lib/python3.11/site-packages/tiktoken/core.py", line 116, in encode
    if match := _special_token_regex(disallowed_special).search(text):
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: expected string or buffer

原因分析 #

从错误栈可以定位到：检索器(retriever)接收的输入不是字符串。你的setup_and_retrieval是RunnableParallel，其中context绑定了retriever，而RunnableParallel会把整个输入字典传给每个分支。当chain2以{'answer': chain1}作为输入时，retriever拿到的不是原始查询字符串，而是这个字典，导致嵌入函数无法处理非字符串输入，触发tiktoken的编码错误。

解决方案 #

修改后代码 #

from source.main.create_db import vectordb 
from langchain.llms import OpenAI
from langchain_core.prompts import ChatPromptTemplate
from dotenv import load_dotenv
import os
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnableParallel, RunnablePassthrough

load_dotenv()
key = os.getenv('OPENAI_API_KEY')
llm = OpenAI()
retriever = vectordb.as_retriever()

# 生成初始回答的chain1
template = '''
You are an assistant specialized in Brazilian public exams, 
mainly focused on tests in the field of law,
and must answer the questions below
{question}
'''
prompt = ChatPromptTemplate.from_template(template)
chain1 = prompt | llm | StrOutputParser()

# 用上下文优化回答的chain2
template2 = '''
Correct and enhance the {answer} using information from the 
{context} and provide more references and support for the newly crafted response.
'''
prompt2 = ChatPromptTemplate.from_template(template2)

# 关键修正：让retriever接收原始question，同时传递chain1的answer
setup_and_retrieval = RunnableParallel(
    context=lambda x: retriever.invoke(x['question']),
    answer=chain1
)

# 调整chain2结构：传入包含question的字典，确保各分支拿到正确输入
chain2 = setup_and_retrieval | prompt2 | llm | StrOutputParser()

# 调用测试
a = chain2.invoke({'question': 'tell me one law'})
print(a)

核心修改点 #

1. 让retriever明确接收原始question字符串，而非整个输入字典
2. 调整chain2的输入格式，传入包含question的字典，确保chain1和retriever都能拿到对应参数
3. 移除未使用的导入，简化代码结构
4. 修正第一个模板的拼写错误（ou → You）

内容的提问来源于stack exchange，提问作者Erick Rocha