gh@.dZddlmZddlmZmZmZddlmZddl m Z m Z ddl m Z ddlmZddlmZdd lmZdd lmZdd lmZdd lmZdd lmZddlmZedddGddeZGddeee e fZ!dS)z+Base classes for LLM-powered router chains.) annotations)AnyOptionalcast) deprecated)AsyncCallbackManagerForChainRunCallbackManagerForChainRun)OutputParserException)BaseLanguageModel)BaseOutputParser)BasePromptTemplate)parse_and_check_json_markdown)model_validator)SelfLLMChain) RouterChainz0.2.12z1.0zUse RunnableLambda to select from multiple prompt templates. See example in API reference: https://api.python.langchain.com/en/latest/chains/langchain.chains.router.llm_router.LLMRouterChain.html)sinceremovalmessageceZdZUdZded< edddZedd Zdfd Z d d!dZ d d"dZ e d#dZ xZS)$LLMRouterChaina A router chain that uses an LLM chain to perform routing. This class is deprecated. See below for a replacement, which offers several benefits, including streaming and batch support. Below is an example implementation: .. code-block:: python from operator import itemgetter from typing import Literal from typing_extensions import TypedDict from langchain_core.output_parsers import StrOutputParser from langchain_core.prompts import ChatPromptTemplate from langchain_core.runnables import RunnableLambda, RunnablePassthrough from langchain_openai import ChatOpenAI llm = ChatOpenAI(model="gpt-4o-mini") prompt_1 = ChatPromptTemplate.from_messages( [ ("system", "You are an expert on animals."), ("human", "{query}"), ] ) prompt_2 = ChatPromptTemplate.from_messages( [ ("system", "You are an expert on vegetables."), ("human", "{query}"), ] ) chain_1 = prompt_1 | llm | StrOutputParser() chain_2 = prompt_2 | llm | StrOutputParser() route_system = "Route the user's query to either the animal or vegetable expert." route_prompt = ChatPromptTemplate.from_messages( [ ("system", route_system), ("human", "{query}"), ] ) class RouteQuery(TypedDict): """Route query to destination.""" destination: Literal["animal", "vegetable"] route_chain = ( route_prompt | llm.with_structured_output(RouteQuery) | itemgetter("destination") ) chain = { "destination": route_chain, # "animal" or "vegetable" "query": lambda x: x["query"], # pass through input query } | RunnableLambda( # if animal, chain_1. otherwise, chain_2. lambda x: chain_1 if x["destination"] == "animal" else chain_2, ) chain.invoke({"query": "what color are carrots"}) r llm_chainafter)modereturnrcJ|jj}|jtd|S)NzLLMRouterChain requires base llm_chain prompt to have an output parser that converts LLM text output to a dictionary with keys 'destination' and 'next_inputs'. Received a prompt with no output parser.)rprompt output_parser ValueError)selfrs _/var/www/FlaskApp/flask-venv/lib/python3.11/site-packages/langchain/chains/router/llm_router.pyvalidate_promptzLLMRouterChain.validate_prompths3&   '   list[str]c|jjS)zTWill be whatever keys the LLM chain prompt expects. :meta private: )r input_keys)r!s r"r'zLLMRouterChain.input_keysts ~((r$outputsdict[str, Any]Nonect|t|dtstdS)N next_inputs)super_validate_outputs isinstancedictr )r!r( __class__s r"r.z LLMRouterChain._validate_outputs|sB !!'***'-0$77    r$Ninputs run_manager$Optional[CallbackManagerForChainRun]c|ptj}|}|jjdd|i|}t t ttf|jj j |}|SN callbacks) r get_noop_manager get_childrpredictrr0strrrrparse)r!r2r3 _run_managerr7 predictionoutputs r"_callzLLMRouterChain._calls #S&@&Q&S&S  **,, +T^+JJiJ6JJ  cN N ! / 5 5j A A   r$)Optional[AsyncCallbackManagerForChainRun]cK|ptj}|}ttt t f|jjdd|i|d{V}|Sr6) r r9r:rr0r<rrapredict_and_parse)r!r2r3r>r7r@s r"_acallzLLMRouterChain._acalls} #S&@&Q&S&S  **,,  cN3$.3RRiR6RR R R R R R R   r$llmr rr kwargsrc 8t||}|dd|i|S)zConvenience constructor.)rFrrr8r)clsrFrrGrs r"from_llmzLLMRouterChain.from_llms1 V444 s11Y1&111r$)rr)rr%)r(r)rr*)N)r2r)r3r4rr))r2r)r3rBrr))rFr rr rGrrr)__name__ __module__ __qualname____doc____annotations__rr#propertyr'r.rArE classmethodrJ __classcell__)r1s@r"rrsAAF+_'"""   #" )))X)=A     $BF     222[22222r$rcFeZdZUdZdZded<eZded<dZded<dd Z d S)RouterOutputParserzrrs11""""""&&&&&&&&&&******<;;;;;<<<<<<::::::555555CCCCCC$$$$$$""""""%%%%%%444444    s B2B2B2B2B2[B2B2B2J)$sCx.9r$