为什么在 MultiPromptChain 上创建路由器提示时不能使用 ChatPromptTemplate 而不是 PromptTemplate？

Question

我是Langchain的新手，最近在学习多路由链。在自己练习的时候，我想知道为什么我们不能使用 ChatPromptTemplate 而不是 PromptTemplate 来获取路由器提示。

Router_temp = MULTI_PROMPT_ROUTER_TEMPLATE.format(destinations = destination_str)
router_prompt = ChatPromptTemplate(template = Router_temp, input_variables=["input"], output_parser= RouterOutputParser())
router_chain = LLMRouterChain.from_llm(llm =llm, prompt = router_prompt)

以下是它抛出的错误： 关键错误：“消息”

但是当我之前在代码中使用 ChatPromptTemplate 来处理类似的情况时，它起作用了。如下所示：

beginner_phy_teach = """You are a Physics teacher who specialises in basic physics concepts. Your name is Feynman, always introduce yourself. Explain these concepts like your audience are five
The topic is as follows\n. {input}"""

advanced_phy_teach = """You are a Physics teacher who specialises in advanced physics concepts. Your name is Oppenheimer, always introduce yourself. Explain these concepts like your audience already know the basics.
The topic is as follows\n. {input}"""

prompt_info = [{"Name": "Beginner Physics Teacher", "Description": "Good for answering questions related to beginner Physics concepts", "Template": beginner_phy_teach
}, {"Name": "Advanced Physics Teacher", "Description": "Good for answering questions related to advanced Physics concepts", "Template": advanced_phy_teach
}]

destination_chains = {}
for p in prompt_info:
    name = p["Name"]
    template = p["Template"]
    chat_prompt = ChatPromptTemplate.from_template(template=template)
    chain = LLMChain(llm=llm, prompt=chat_prompt)
    destination_chains[name] = chain

Answer 1

使用 ChatPromptTemplate 创建模板，然后将模板传递给大语言模型

llm = ChatOpenAI(temperature=0, openai_api_key=openai_api_key)      

# Create a chat prompt template
prompt_template = ChatPromptTemplate.from_messages(
    [
        ("system", "Your an smart ai agent."),
        ("human", "Respond to question: {question}")
    ]
)

# Insert a question into the template and call the model
full_prompt = prompt_template.format_messages(question='Why is the sky blue?')
llm(full_prompt)