How to merge consecutive messages of the same type

Certain models do not support passing in consecutive messages of the same type (a.k.a. "runs" of the same message type).

The merge_message_runs utility makes it easy to merge consecutive messages of the same type.

Setup

%pip install -qU langchain-core langchain-anthropic

Basic usage

from langchain_core.messages import (
    AIMessage,
    HumanMessage,
    SystemMessage,
    merge_message_runs,
)

messages = [
    SystemMessage("you're a good assistant."),
    SystemMessage("you always respond with a joke."),
    HumanMessage([{"type": "text", "text": "i wonder why it's called langchain"}]),
    HumanMessage("and who is harrison chasing anyways"),
    AIMessage(
        'Well, I guess they thought "WordRope" and "SentenceString" just didn\'t have the same ring to it!'
    ),
    AIMessage("Why, he's probably chasing after the last cup of coffee in the office!"),
]

merged = merge_message_runs(messages)
print("\n\n".join([repr(x) for x in merged]))

API Reference:AIMessage | HumanMessage | SystemMessage | merge_message_runs

SystemMessage(content="you're a good assistant.\nyou always respond with a joke.", additional_kwargs={}, response_metadata={})

HumanMessage(content=[{'type': 'text', 'text': "i wonder why it's called langchain"}, 'and who is harrison chasing anyways'], additional_kwargs={}, response_metadata={})

AIMessage(content='Well, I guess they thought "WordRope" and "SentenceString" just didn\'t have the same ring to it!\nWhy, he\'s probably chasing after the last cup of coffee in the office!', additional_kwargs={}, response_metadata={})

Notice that if the contents of one of the messages to merge is a list of content blocks then the merged message will have a list of content blocks. And if both messages to merge have string contents then those are concatenated with a newline character.

Chaining

merge_message_runs can be used in an imperatively (like above) or declaratively, making it easy to compose with other components in a chain:

%pip install -qU langchain-anthropic
from langchain_anthropic import ChatAnthropic

llm = ChatAnthropic(model="claude-3-sonnet-20240229", temperature=0)
# Notice we don't pass in messages. This creates
# a RunnableLambda that takes messages as input
merger = merge_message_runs()
chain = merger | llm
chain.invoke(messages)

API Reference:ChatAnthropic

Note: you may need to restart the kernel to use updated packages.

AIMessage(content=[], additional_kwargs={}, response_metadata={'id': 'msg_01KNGUMTuzBVfwNouLDpUMwf', 'model': 'claude-3-sonnet-20240229', 'stop_reason': 'end_turn', 'stop_sequence': None, 'usage': {'input_tokens': 84, 'output_tokens': 3}}, id='run-b908b198-9c24-450b-9749-9d4a8182937b-0', usage_metadata={'input_tokens': 84, 'output_tokens': 3, 'total_tokens': 87})

Looking at the LangSmith trace we can see that before the messages are passed to the model they are merged: https://smith.langchain.com/public/ab558677-cac9-4c59-9066-1ecce5bcd87c/r

Looking at just the merger, we can see that it's a Runnable object that can be invoked like all Runnables:

merger.invoke(messages)

[SystemMessage(content="you're a good assistant.\nyou always respond with a joke.", additional_kwargs={}, response_metadata={}),
 HumanMessage(content=[{'type': 'text', 'text': "i wonder why it's called langchain"}, 'and who is harrison chasing anyways'], additional_kwargs={}, response_metadata={}),
 AIMessage(content='Well, I guess they thought "WordRope" and "SentenceString" just didn\'t have the same ring to it!\nWhy, he\'s probably chasing after the last cup of coffee in the office!', additional_kwargs={}, response_metadata={})]

merge_message_runs can also be placed after a prompt:

from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate(
    [
        ("system", "You're great a {skill}"),
        ("system", "You're also great at explaining things"),
        ("human", "{query}"),
    ]
)
chain = prompt | merger | llm
chain.invoke({"skill": "math", "query": "what's the definition of a convergent series"})

API Reference:ChatPromptTemplate

AIMessage(content='A convergent series is an infinite series whose partial sums approach a finite value as more terms are added. In other words, the sequence of partial sums has a limit.\n\nMore formally, an infinite series Σ an (where an are the terms of the series) is said to be convergent if the sequence of partial sums:\n\nS1 = a1\nS2 = a1 + a2  \nS3 = a1 + a2 + a3\n...\nSn = a1 + a2 + a3 + ... + an\n...\n\nconverges to some finite number S as n goes to infinity. We write:\n\nlim n→∞ Sn = S\n\nThe finite number S is called the sum of the convergent infinite series.\n\nIf the sequence of partial sums does not approach any finite limit, the infinite series is said to be divergent.\n\nSome key properties:\n- A series converges if and only if the sequence of its partial sums is a Cauchy sequence.\n- Absolute/conditional convergence criteria help determine if a given series converges.\n- Convergent series have many important applications in mathematics, physics, engineering etc.', additional_kwargs={}, response_metadata={'id': 'msg_01MfV6y2hep7ZNvDz24A36U4', 'model': 'claude-3-sonnet-20240229', 'stop_reason': 'end_turn', 'stop_sequence': None, 'usage': {'input_tokens': 29, 'output_tokens': 267}}, id='run-9d925f58-021e-4bd0-94fc-f8f5e91010a4-0', usage_metadata={'input_tokens': 29, 'output_tokens': 267, 'total_tokens': 296})

LangSmith Trace: https://smith.langchain.com/public/432150b6-9909-40a7-8ae7-944b7e657438/r/f4ad5fb2-4d38-42a6-b780-25f62617d53f

API reference

For a complete description of all arguments head to the API reference: https://python.langchain.com/api_reference/core/messages/langchain_core.messages.utils.merge_message_runs.html