U hp#@sdZddlmZddlmZmZmZmZddlm Z ddl m Z ddl m Z mZddlmZddlmZmZdd lmZdd lmZd d d dZGdddeZdS)zLCombine documents by doing a first pass and then refining on more documents.) annotations)AnyDictListTuple) Callbacks)Document)BasePromptTemplateformat_documentPromptTemplate)Fieldroot_validator)BaseCombineDocumentsChain)LLMChainr returncCstdgddS)N page_contentz{page_content})input_variablestemplater rrM/tmp/pip-unpacked-wheel-bo69hh5q/langchain/chains/combine_documents/refine.py_get_default_document_promptsrcs eZdZUdZded<ded<ded<ded<eedZd ed <d Zd ed <e ddfdd Z GdddZ e dddddddZ e dddddddZd3dddd d!d"d#Zd4dddd d!d$d%Zddd d&d'd(Zd)dd*d+d,d-Zddd*d.d/d0Ze ddd1d2ZZS)5RefineDocumentsChaina& Combine documents by doing a first pass and then refining on more documents. This algorithm first calls `initial_llm_chain` on the first document, passing that first document in with the variable name `document_variable_name`, and produces a new variable with the variable name `initial_response_name`. Then, it loops over every remaining document. This is called the "refine" step. It calls `refine_llm_chain`, passing in that document with the variable name `document_variable_name` as well as the previous response with the variable name `initial_response_name`. Example: .. code-block:: python from langchain.chains import RefineDocumentsChain, LLMChain from langchain_core.prompts import PromptTemplate from langchain_community.llms import OpenAI # This controls how each document will be formatted. Specifically, # it will be passed to `format_document` - see that function for more # details. document_prompt = PromptTemplate( input_variables=["page_content"], template="{page_content}" ) document_variable_name = "context" llm = OpenAI() # The prompt here should take as an input variable the # `document_variable_name` prompt = PromptTemplate.from_template( "Summarize this content: {context}" ) initial_llm_chain = LLMChain(llm=llm, prompt=prompt) initial_response_name = "prev_response" # The prompt here should take as an input variable the # `document_variable_name` as well as `initial_response_name` prompt_refine = PromptTemplate.from_template( "Here's your first summary: {prev_response}. " "Now add to it based on the following context: {context}" ) refine_llm_chain = LLMChain(llm=llm, prompt=prompt_refine) chain = RefineDocumentsChain( initial_llm_chain=initial_llm_chain, refine_llm_chain=refine_llm_chain, document_prompt=document_prompt, document_variable_name=document_variable_name, initial_response_name=initial_response_name, ) rinitial_llm_chainrefine_llm_chainstrdocument_variable_nameinitial_response_name)default_factoryr document_promptFboolreturn_intermediate_stepsz List[str]rcstj}|jr|dg}|S)z2Expect input key. :meta private: intermediate_steps)super output_keysr")selfZ _output_keys __class__rrr%Zs z RefineDocumentsChain.output_keysc@seZdZdZdZdS)zRefineDocumentsChain.ConfigTZforbidN)__name__ __module__ __qualname__Zarbitrary_types_allowedextrarrrrConfigesr-T)prer)valuesrcCsd|kr|d|d<|d=|S)zFor backwards compatibility.Zreturn_refine_stepsr"r)clsr/rrrget_return_intermediate_stepsis z2RefineDocumentsChain.get_return_intermediate_stepscCspd|krtd|djj}d|krHt|dkr>|d|d<qltdn$|d|krltd|dd||S) z4Get default document variable name, if not provided.rz"initial_llm_chain must be providedrrzWdocument_variable_name must be provided if there are multiple llm_chain input_variableszdocument_variable_name z- was not found in llm_chain input_variables: ) ValueErrorpromptrlen)r0r/Zllm_chain_variablesrrr"get_default_document_variable_nameqs   z7RefineDocumentsChain.get_default_document_variable_nameNzList[Document]rrzTuple[str, dict])docs callbackskwargsrc Ks||j|f|}|jjfd|i|}|g}|ddD]8}|||}||}|jjfd|i|}||q6|||S)aCombine by mapping first chain over all, then stuffing into final chain. Args: docs: List of documents to combine callbacks: Callbacks to be passed through **kwargs: additional parameters to be passed to LLM calls (like other input variables besides the documents) Returns: The first element returned is the single string output. The second element returned is a dictionary of other keys to return. r8r2N)_construct_initial_inputsrZpredict_construct_refine_inputsrappend_construct_result r&r7r8r9inputsres refine_stepsdoc base_inputsrrr combine_docss  z!RefineDocumentsChain.combine_docsc s|j|f|}|jjfd|i|IdH}|g}|ddD]>}|||}||}|jjfd|i|IdH}||q<|||S)aAsync combine by mapping a first chain over all, then stuffing into a final chain. Args: docs: List of documents to combine callbacks: Callbacks to be passed through **kwargs: additional parameters to be passed to LLM calls (like other input variables besides the documents) Returns: The first element returned is the single string output. The second element returned is a dictionary of other keys to return. r8Nr2)r:rZapredictr;rr<r=r>rrr acombine_docss  z"RefineDocumentsChain.acombine_docs)rAr@rcCs|jrd|i}ni}||fS)Nr#)r")r&rAr@Zextra_return_dictrrrr=s z&RefineDocumentsChain._construct_resultrzDict[str, Any])rBr@rcCs|jt||j|j|iS)N)rr r r)r&rBr@rrrr;s  z-RefineDocumentsChain._construct_refine_inputs)r7r9rc sTd|dji|djfdd|jjD}|j|jjf|i}||}|S)Nrrcsi|]}||qSrr).0kZ base_inforr szBRefineDocumentsChain._construct_initial_inputs..)rupdatemetadatar rrformat)r&r7r9Z document_inforCr?rrHrr:s z.RefineDocumentsChain._construct_initial_inputscCsdS)NZrefine_documents_chainr)r&rrr _chain_typesz RefineDocumentsChain._chain_type)N)N)r)r*r+__doc____annotations__r rr r"propertyr%r-rr1r6rDrEr=r;r:rM __classcell__rrr'rrs2 2   rN)rN __future__rtypingrrrrZlangchain_core.callbacksrZlangchain_core.documentsrZlangchain_core.promptsr r Zlangchain_core.prompts.promptr Zlangchain_core.pydantic_v1r rZ'langchain.chains.combine_documents.baserZlangchain.chains.llmrrrrrrrs