Source code for langchain.chains.conversational_retrieval.base
"""Chain for chatting with a vector database."""from__future__importannotationsimportinspectimportwarningsfromabcimportabstractmethodfrompathlibimportPathfromtypingimportAny,Callable,Dict,List,Optional,Tuple,Type,Unionfromlangchain_core._apiimportdeprecatedfromlangchain_core.callbacksimport(AsyncCallbackManagerForChainRun,CallbackManagerForChainRun,Callbacks,)fromlangchain_core.documentsimportDocumentfromlangchain_core.language_modelsimportBaseLanguageModelfromlangchain_core.messagesimportBaseMessagefromlangchain_core.promptsimportBasePromptTemplatefromlangchain_core.retrieversimportBaseRetrieverfromlangchain_core.runnablesimportRunnableConfigfromlangchain_core.vectorstoresimportVectorStorefrompydanticimportBaseModel,ConfigDict,Field,model_validatorfromlangchain.chains.baseimportChainfromlangchain.chains.combine_documents.baseimportBaseCombineDocumentsChainfromlangchain.chains.combine_documents.stuffimportStuffDocumentsChainfromlangchain.chains.conversational_retrieval.promptsimportCONDENSE_QUESTION_PROMPTfromlangchain.chains.llmimportLLMChainfromlangchain.chains.question_answeringimportload_qa_chain# Depending on the memory type and configuration, the chat history format may differ.# This needs to be consolidated.CHAT_TURN_TYPE=Union[Tuple[str,str],BaseMessage]_ROLE_MAP={"human":"Human: ","ai":"Assistant: "}def_get_chat_history(chat_history:List[CHAT_TURN_TYPE])->str:buffer=""fordialogue_turninchat_history:ifisinstance(dialogue_turn,BaseMessage):iflen(dialogue_turn.content)>0:role_prefix=_ROLE_MAP.get(dialogue_turn.type,f"{dialogue_turn.type}: ")buffer+=f"\n{role_prefix}{dialogue_turn.content}"elifisinstance(dialogue_turn,tuple):human="Human: "+dialogue_turn[0]ai="Assistant: "+dialogue_turn[1]buffer+="\n"+"\n".join([human,ai])else:raiseValueError(f"Unsupported chat history format: {type(dialogue_turn)}."f" Full chat history: {chat_history} ")returnbuffer
[docs]classInputType(BaseModel):"""Input type for ConversationalRetrievalChain."""question:str"""The question to answer."""chat_history:List[CHAT_TURN_TYPE]=Field(default_factory=list)"""The chat history to use for retrieval."""
[docs]classBaseConversationalRetrievalChain(Chain):"""Chain for chatting with an index."""combine_docs_chain:BaseCombineDocumentsChain"""The chain used to combine any retrieved documents."""question_generator:LLMChain"""The chain used to generate a new question for the sake of retrieval. This chain will take in the current question (with variable `question`) and any chat history (with variable `chat_history`) and will produce a new standalone question to be used later on."""output_key:str="answer""""The output key to return the final answer of this chain in."""rephrase_question:bool=True"""Whether or not to pass the new generated question to the combine_docs_chain. If True, will pass the new generated question along. If False, will only use the new generated question for retrieval and pass the original question along to the combine_docs_chain."""return_source_documents:bool=False"""Return the retrieved source documents as part of the final result."""return_generated_question:bool=False"""Return the generated question as part of the final result."""get_chat_history:Optional[Callable[[List[CHAT_TURN_TYPE]],str]]=None"""An optional function to get a string of the chat history. If None is provided, will use a default."""response_if_no_docs_found:Optional[str]=None"""If specified, the chain will return a fixed response if no docs are found for the question. """model_config=ConfigDict(populate_by_name=True,arbitrary_types_allowed=True,extra="forbid",)@propertydefinput_keys(self)->List[str]:"""Input keys."""return["question","chat_history"]defget_input_schema(self,config:Optional[RunnableConfig]=None)->Type[BaseModel]:returnInputType@propertydefoutput_keys(self)->List[str]:"""Return the output keys. :meta private: """_output_keys=[self.output_key]ifself.return_source_documents:_output_keys=_output_keys+["source_documents"]ifself.return_generated_question:_output_keys=_output_keys+["generated_question"]return_output_keys@abstractmethoddef_get_docs(self,question:str,inputs:Dict[str,Any],*,run_manager:CallbackManagerForChainRun,)->List[Document]:"""Get docs."""def_call(self,inputs:Dict[str,Any],run_manager:Optional[CallbackManagerForChainRun]=None,)->Dict[str,Any]:_run_manager=run_managerorCallbackManagerForChainRun.get_noop_manager()question=inputs["question"]get_chat_history=self.get_chat_historyor_get_chat_historychat_history_str=get_chat_history(inputs["chat_history"])ifchat_history_str:callbacks=_run_manager.get_child()new_question=self.question_generator.run(question=question,chat_history=chat_history_str,callbacks=callbacks)else:new_question=questionaccepts_run_manager=("run_manager"ininspect.signature(self._get_docs).parameters)ifaccepts_run_manager:docs=self._get_docs(new_question,inputs,run_manager=_run_manager)else:docs=self._get_docs(new_question,inputs)# type: ignore[call-arg]output:Dict[str,Any]={}ifself.response_if_no_docs_foundisnotNoneandlen(docs)==0:output[self.output_key]=self.response_if_no_docs_foundelse:new_inputs=inputs.copy()ifself.rephrase_question:new_inputs["question"]=new_questionnew_inputs["chat_history"]=chat_history_stranswer=self.combine_docs_chain.run(input_documents=docs,callbacks=_run_manager.get_child(),**new_inputs)output[self.output_key]=answerifself.return_source_documents:output["source_documents"]=docsifself.return_generated_question:output["generated_question"]=new_questionreturnoutput@abstractmethodasyncdef_aget_docs(self,question:str,inputs:Dict[str,Any],*,run_manager:AsyncCallbackManagerForChainRun,)->List[Document]:"""Get docs."""asyncdef_acall(self,inputs:Dict[str,Any],run_manager:Optional[AsyncCallbackManagerForChainRun]=None,)->Dict[str,Any]:_run_manager=run_managerorAsyncCallbackManagerForChainRun.get_noop_manager()question=inputs["question"]get_chat_history=self.get_chat_historyor_get_chat_historychat_history_str=get_chat_history(inputs["chat_history"])ifchat_history_str:callbacks=_run_manager.get_child()new_question=awaitself.question_generator.arun(question=question,chat_history=chat_history_str,callbacks=callbacks)else:new_question=questionaccepts_run_manager=("run_manager"ininspect.signature(self._aget_docs).parameters)ifaccepts_run_manager:docs=awaitself._aget_docs(new_question,inputs,run_manager=_run_manager)else:docs=awaitself._aget_docs(new_question,inputs)# type: ignore[call-arg]output:Dict[str,Any]={}ifself.response_if_no_docs_foundisnotNoneandlen(docs)==0:output[self.output_key]=self.response_if_no_docs_foundelse:new_inputs=inputs.copy()ifself.rephrase_question:new_inputs["question"]=new_questionnew_inputs["chat_history"]=chat_history_stranswer=awaitself.combine_docs_chain.arun(input_documents=docs,callbacks=_run_manager.get_child(),**new_inputs)output[self.output_key]=answerifself.return_source_documents:output["source_documents"]=docsifself.return_generated_question:output["generated_question"]=new_questionreturnoutput
[docs]defsave(self,file_path:Union[Path,str])->None:ifself.get_chat_history:raiseValueError("Chain not saveable when `get_chat_history` is not None.")super().save(file_path)
[docs]@deprecated(since="0.1.17",alternative=("create_history_aware_retriever together with create_retrieval_chain ""(see example in docstring)"),removal="1.0",)classConversationalRetrievalChain(BaseConversationalRetrievalChain):"""Chain for having a conversation based on retrieved documents. This class is deprecated. See below for an example implementation using `create_retrieval_chain`. Additional walkthroughs can be found at https://python.langchain.com/docs/use_cases/question_answering/chat_history .. code-block:: python from langchain.chains import ( create_history_aware_retriever, create_retrieval_chain, ) from langchain.chains.combine_documents import create_stuff_documents_chain from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder from langchain_openai import ChatOpenAI retriever = ... # Your retriever llm = ChatOpenAI() # Contextualize question contextualize_q_system_prompt = ( "Given a chat history and the latest user question " "which might reference context in the chat history, " "formulate a standalone question which can be understood " "without the chat history. Do NOT answer the question, just " "reformulate it if needed and otherwise return it as is." ) contextualize_q_prompt = ChatPromptTemplate.from_messages( [ ("system", contextualize_q_system_prompt), MessagesPlaceholder("chat_history"), ("human", "{input}"), ] ) history_aware_retriever = create_history_aware_retriever( llm, retriever, contextualize_q_prompt ) # Answer question qa_system_prompt = ( "You are an assistant for question-answering tasks. Use " "the following pieces of retrieved context to answer the " "question. If you don't know the answer, just say that you " "don't know. Use three sentences maximum and keep the answer " "concise." "\n\n" "{context}" ) qa_prompt = ChatPromptTemplate.from_messages( [ ("system", qa_system_prompt), MessagesPlaceholder("chat_history"), ("human", "{input}"), ] ) # Below we use create_stuff_documents_chain to feed all retrieved context # into the LLM. Note that we can also use StuffDocumentsChain and other # instances of BaseCombineDocumentsChain. question_answer_chain = create_stuff_documents_chain(llm, qa_prompt) rag_chain = create_retrieval_chain( history_aware_retriever, question_answer_chain ) # Usage: chat_history = [] # Collect chat history here (a sequence of messages) rag_chain.invoke({"input": query, "chat_history": chat_history}) This chain takes in chat history (a list of messages) and new questions, and then returns an answer to that question. The algorithm for this chain consists of three parts: 1. Use the chat history and the new question to create a "standalone question". This is done so that this question can be passed into the retrieval step to fetch relevant documents. If only the new question was passed in, then relevant context may be lacking. If the whole conversation was passed into retrieval, there may be unnecessary information there that would distract from retrieval. 2. This new question is passed to the retriever and relevant documents are returned. 3. The retrieved documents are passed to an LLM along with either the new question (default behavior) or the original question and chat history to generate a final response. Example: .. code-block:: python from langchain.chains import ( StuffDocumentsChain, LLMChain, ConversationalRetrievalChain ) from langchain_core.prompts import PromptTemplate from langchain_community.llms import OpenAI combine_docs_chain = StuffDocumentsChain(...) vectorstore = ... retriever = vectorstore.as_retriever() # This controls how the standalone question is generated. # Should take `chat_history` and `question` as input variables. template = ( "Combine the chat history and follow up question into " "a standalone question. Chat History: {chat_history}" "Follow up question: {question}" ) prompt = PromptTemplate.from_template(template) llm = OpenAI() question_generator_chain = LLMChain(llm=llm, prompt=prompt) chain = ConversationalRetrievalChain( combine_docs_chain=combine_docs_chain, retriever=retriever, question_generator=question_generator_chain, ) """retriever:BaseRetriever"""Retriever to use to fetch documents."""max_tokens_limit:Optional[int]=None"""If set, enforces that the documents returned are less than this limit. This is only enforced if `combine_docs_chain` is of type StuffDocumentsChain."""def_reduce_tokens_below_limit(self,docs:List[Document])->List[Document]:num_docs=len(docs)ifself.max_tokens_limitandisinstance(self.combine_docs_chain,StuffDocumentsChain):tokens=[self.combine_docs_chain.llm_chain._get_num_tokens(doc.page_content)fordocindocs]token_count=sum(tokens[:num_docs])whiletoken_count>self.max_tokens_limit:num_docs-=1token_count-=tokens[num_docs]returndocs[:num_docs]def_get_docs(self,question:str,inputs:Dict[str,Any],*,run_manager:CallbackManagerForChainRun,)->List[Document]:"""Get docs."""docs=self.retriever.invoke(question,config={"callbacks":run_manager.get_child()})returnself._reduce_tokens_below_limit(docs)asyncdef_aget_docs(self,question:str,inputs:Dict[str,Any],*,run_manager:AsyncCallbackManagerForChainRun,)->List[Document]:"""Get docs."""docs=awaitself.retriever.ainvoke(question,config={"callbacks":run_manager.get_child()})returnself._reduce_tokens_below_limit(docs)
[docs]@classmethoddeffrom_llm(cls,llm:BaseLanguageModel,retriever:BaseRetriever,condense_question_prompt:BasePromptTemplate=CONDENSE_QUESTION_PROMPT,chain_type:str="stuff",verbose:bool=False,condense_question_llm:Optional[BaseLanguageModel]=None,combine_docs_chain_kwargs:Optional[Dict]=None,callbacks:Callbacks=None,**kwargs:Any,)->BaseConversationalRetrievalChain:"""Convenience method to load chain from LLM and retriever. This provides some logic to create the `question_generator` chain as well as the combine_docs_chain. Args: llm: The default language model to use at every part of this chain (eg in both the question generation and the answering) retriever: The retriever to use to fetch relevant documents from. condense_question_prompt: The prompt to use to condense the chat history and new question into a standalone question. chain_type: The chain type to use to create the combine_docs_chain, will be sent to `load_qa_chain`. verbose: Verbosity flag for logging to stdout. condense_question_llm: The language model to use for condensing the chat history and new question into a standalone question. If none is provided, will default to `llm`. combine_docs_chain_kwargs: Parameters to pass as kwargs to `load_qa_chain` when constructing the combine_docs_chain. callbacks: Callbacks to pass to all subchains. kwargs: Additional parameters to pass when initializing ConversationalRetrievalChain """combine_docs_chain_kwargs=combine_docs_chain_kwargsor{}doc_chain=load_qa_chain(llm,chain_type=chain_type,verbose=verbose,callbacks=callbacks,**combine_docs_chain_kwargs,)_llm=condense_question_llmorllmcondense_question_chain=LLMChain(llm=_llm,prompt=condense_question_prompt,verbose=verbose,callbacks=callbacks,)returncls(retriever=retriever,combine_docs_chain=doc_chain,question_generator=condense_question_chain,callbacks=callbacks,**kwargs,)
[docs]classChatVectorDBChain(BaseConversationalRetrievalChain):"""Chain for chatting with a vector database."""vectorstore:VectorStore=Field(alias="vectorstore")top_k_docs_for_context:int=4search_kwargs:dict=Field(default_factory=dict)@propertydef_chain_type(self)->str:return"chat-vector-db"@model_validator(mode="before")@classmethoddefraise_deprecation(cls,values:Dict)->Any:warnings.warn("`ChatVectorDBChain` is deprecated - ""please use `from langchain.chains import ConversationalRetrievalChain`")returnvaluesdef_get_docs(self,question:str,inputs:Dict[str,Any],*,run_manager:CallbackManagerForChainRun,)->List[Document]:"""Get docs."""vectordbkwargs=inputs.get("vectordbkwargs",{})full_kwargs={**self.search_kwargs,**vectordbkwargs}returnself.vectorstore.similarity_search(question,k=self.top_k_docs_for_context,**full_kwargs)asyncdef_aget_docs(self,question:str,inputs:Dict[str,Any],*,run_manager:AsyncCallbackManagerForChainRun,)->List[Document]:"""Get docs."""raiseNotImplementedError("ChatVectorDBChain does not support async")
[docs]@classmethoddeffrom_llm(cls,llm:BaseLanguageModel,vectorstore:VectorStore,condense_question_prompt:BasePromptTemplate=CONDENSE_QUESTION_PROMPT,chain_type:str="stuff",combine_docs_chain_kwargs:Optional[Dict]=None,callbacks:Callbacks=None,**kwargs:Any,)->BaseConversationalRetrievalChain:"""Load chain from LLM."""combine_docs_chain_kwargs=combine_docs_chain_kwargsor{}doc_chain=load_qa_chain(llm,chain_type=chain_type,callbacks=callbacks,**combine_docs_chain_kwargs,)condense_question_chain=LLMChain(llm=llm,prompt=condense_question_prompt,callbacks=callbacks)returncls(vectorstore=vectorstore,combine_docs_chain=doc_chain,question_generator=condense_question_chain,callbacks=callbacks,**kwargs,)