Chained Prompt Template

Sometimes, you may want to append prompt templates together. For example:

• When you want to reuse the same chunk of templating in multiple places

• When you are running a chain that depends on prior context

In that case, you can use the ChainedPromptTemplate to chain these templates together. (Note the possibility of intermingling PromptTemplate’s with plain strings.)

from langchain.prompts.prompt import PromptTemplate
from langchain_contrib.prompts import ChainedPromptTemplate

template = ChainedPromptTemplate([
    PromptTemplate.from_template("You have access to {tools}."),
    "Your objective is to answer the question: {question}?",
    "What is your next action?",
], joiner="\n\n")
print(template.format(tools="Search", question="how high is Everest"))

You have access to Search.

Your objective is to answer the question: how high is Everest?

What is your next action?

Chaining chat templates

You can also chain arbitrary chat prompt templates or message prompt templates together. Plain strings are intepreted as Human messages.

from langchain.prompts.chat import ChatPromptTemplate, SystemMessagePromptTemplate

template = ChainedPromptTemplate([
    SystemMessagePromptTemplate.from_template("You have access to {tools}."),
    ChatPromptTemplate.from_messages([
        SystemMessagePromptTemplate.from_template("Your objective is to answer human questions."),
    ]),
    "Tell me: {question}?",
])
template.format_prompt(tools="Search", question="how high is Everest").to_messages()

[SystemMessage(content='You have access to Search.', additional_kwargs={}),
 SystemMessage(content='Your objective is to answer human questions.', additional_kwargs={}),
 HumanMessage(content='Tell me: how high is Everest?', additional_kwargs={})]

Chained Prompt Values

If you already have PromptValue’s instead of PromptTemplate’s and just want to chain these values up, you can create a ChainedPromptValue. The values can be a mix of StringPromptValue and ChatPromptValue.

from langchain.prompts.base import StringPromptValue
from langchain.prompts.chat import ChatPromptValue
from langchain.schema import SystemMessage, AIMessage
from langchain_contrib.prompts import ChainedPromptValue

value = ChainedPromptValue(
    joiner="\n",
    subvalues=[
        ChatPromptValue(
            messages=[
                SystemMessage(content="You have access to Search."),
                AIMessage(content="What can I help with?"),
            ]
        ),
        StringPromptValue(text="What is langchain-contrib?"),
    ],
)

print(value.to_string())

System: You have access to Search.
AI: What can I help with?
What is langchain-contrib?

Chat message serialization is also supported.

value.to_messages()

[SystemMessage(content='You have access to Search.', additional_kwargs={}),
 AIMessage(content='What can I help with?', additional_kwargs={}),
 HumanMessage(content='What is langchain-contrib?', additional_kwargs={})]

Prefixing prompt templates

If you specifically want to just prepend a prefix to the current prompt template, as opposed to the much more general capabilities that ChainedPromptTemplate offers, then head on to the next page to learn about using PrefixedTemplate to do just that.