如何修剪消息

先决条件

本指南假设您熟悉以下概念

• 消息
• 聊天模型
• 链式
• 聊天历史

本指南中的方法还需要langchain-core>=0.2.9。

所有模型都有有限的上下文窗口，这意味着它们可以接受的令牌输入数量是有限的。 如果您有非常长的消息或累积了长消息历史的链/代理，则需要管理传递给模型的的消息长度。

trim_messages 可用于将聊天历史记录的大小减少到指定的令牌计数或指定的消息计数。

如果将修剪后的聊天历史记录直接传递回聊天模型，则修剪后的聊天历史记录应满足以下属性

1. 生成的聊天历史记录应为有效的。 通常，这意味着应满足以下属性
   • 聊天历史记录以 (1) HumanMessage 或 (2) SystemMessage 后跟 HumanMessage 开头。
   • 聊天历史记录以 HumanMessage 或 ToolMessage 结尾。
   • ToolMessage 只能出现在涉及工具调用的 AIMessage 之后。这可以通过设置 start_on="human" 和 ends_on=("human", "tool") 来实现。
2. 它包括最近的消息，并删除聊天历史记录中的旧消息。这可以通过设置 strategy="last" 来实现。
3. 通常，新的聊天历史记录应包括原始聊天历史记录中存在的 SystemMessage，因为 SystemMessage 包含对聊天模型的特殊指令。 如果存在，SystemMessage 几乎总是历史记录中的第一条消息。这可以通过设置 include_system=True 来实现。

基于令牌计数修剪

在这里，我们将根据令牌计数修剪聊天历史记录。 修剪后的聊天历史记录将生成一个包含 SystemMessage 的有效聊天历史记录。

为了保留最近的消息，我们设置 strategy="last"。 我们还将设置 include_system=True 以包含 SystemMessage，并设置 start_on="human" 以确保生成的聊天历史记录有效。

当基于令牌计数使用 trim_messages 时，这是一个很好的默认配置。 请记住根据您的用例调整 token_counter 和 max_tokens。

请注意，对于我们的 token_counter，我们可以传入一个函数（更多内容见下文）或一个语言模型（因为语言模型具有消息令牌计数方法）。 当您修剪消息以适应特定模型的上下文窗口时，传入模型是有意义的

pip install -qU langchain-openai

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

from langchain_core.messages import (
    AIMessage,
    HumanMessage,
    SystemMessage,
    ToolMessage,
    trim_messages,
)
from langchain_openai import ChatOpenAI

messages = [
    SystemMessage("you're a good assistant, you always respond with a joke."),
    HumanMessage("i wonder why it's called langchain"),
    AIMessage(
        'Well, I guess they thought "WordRope" and "SentenceString" just didn\'t have the same ring to it!'
    ),
    HumanMessage("and who is harrison chasing anyways"),
    AIMessage(
        "Hmmm let me think.\n\nWhy, he's probably chasing after the last cup of coffee in the office!"
    ),
    HumanMessage("what do you call a speechless parrot"),
]


trim_messages(
    messages,
    # Keep the last <= n_count tokens of the messages.
    strategy="last",
    # Remember to adjust based on your model
    # or else pass a custom token_encoder
    token_counter=ChatOpenAI(model="gpt-4o"),
    # Most chat models expect that chat history starts with either:
    # (1) a HumanMessage or
    # (2) a SystemMessage followed by a HumanMessage
    # Remember to adjust based on the desired conversation
    # length
    max_tokens=45,
    # Most chat models expect that chat history starts with either:
    # (1) a HumanMessage or
    # (2) a SystemMessage followed by a HumanMessage
    start_on="human",
    # Most chat models expect that chat history ends with either:
    # (1) a HumanMessage or
    # (2) a ToolMessage
    end_on=("human", "tool"),
    # Usually, we want to keep the SystemMessage
    # if it's present in the original history.
    # The SystemMessage has special instructions for the model.
    include_system=True,
    allow_partial=False,
)

API 参考:AIMessage | HumanMessage | SystemMessage | ToolMessage | trim_messages | ChatOpenAI

[SystemMessage(content="you're a good assistant, you always respond with a joke.", additional_kwargs={}, response_metadata={}),
 HumanMessage(content='what do you call a speechless parrot', additional_kwargs={}, response_metadata={})]

基于消息计数修剪

或者，我们可以通过设置 token_counter=len 来根据消息计数修剪聊天历史记录。 在这种情况下，每条消息将计为一个令牌，而 max_tokens 将控制最大消息数。

当基于消息计数使用 trim_messages 时，这是一个很好的默认配置。请记住根据您的用例调整 max_tokens。

trim_messages(
    messages,
    # Keep the last <= n_count tokens of the messages.
    strategy="last",
    token_counter=len,
    # When token_counter=len, each message
    # will be counted as a single token.
    # Remember to adjust for your use case
    max_tokens=5,
    # Most chat models expect that chat history starts with either:
    # (1) a HumanMessage or
    # (2) a SystemMessage followed by a HumanMessage
    start_on="human",
    # Most chat models expect that chat history ends with either:
    # (1) a HumanMessage or
    # (2) a ToolMessage
    end_on=("human", "tool"),
    # Usually, we want to keep the SystemMessage
    # if it's present in the original history.
    # The SystemMessage has special instructions for the model.
    include_system=True,
)

[SystemMessage(content="you're a good assistant, you always respond with a joke.", additional_kwargs={}, response_metadata={}),
 HumanMessage(content='and who is harrison chasing anyways', additional_kwargs={}, response_metadata={}),
 AIMessage(content="Hmmm let me think.\n\nWhy, he's probably chasing after the last cup of coffee in the office!", additional_kwargs={}, response_metadata={}),
 HumanMessage(content='what do you call a speechless parrot', additional_kwargs={}, response_metadata={})]

高级用法

您可以将 trim_message 用作构建块来创建更复杂的处理逻辑。

如果我们想允许拆分消息的内容，我们可以指定 allow_partial=True

trim_messages(
    messages,
    max_tokens=56,
    strategy="last",
    token_counter=ChatOpenAI(model="gpt-4o"),
    include_system=True,
    allow_partial=True,
)

[SystemMessage(content="you're a good assistant, you always respond with a joke.", additional_kwargs={}, response_metadata={}),
 AIMessage(content="\nWhy, he's probably chasing after the last cup of coffee in the office!", additional_kwargs={}, response_metadata={}),
 HumanMessage(content='what do you call a speechless parrot', additional_kwargs={}, response_metadata={})]

默认情况下，不会包含 SystemMessage，因此您可以通过设置 include_system=False 或删除 include_system 参数来删除它。

trim_messages(
    messages,
    max_tokens=45,
    strategy="last",
    token_counter=ChatOpenAI(model="gpt-4o"),
)

[AIMessage(content="Hmmm let me think.\n\nWhy, he's probably chasing after the last cup of coffee in the office!", additional_kwargs={}, response_metadata={}),
 HumanMessage(content='what do you call a speechless parrot', additional_kwargs={}, response_metadata={})]

我们可以通过指定 strategy="first" 来执行获取第一个 max_tokens 的翻转操作

trim_messages(
    messages,
    max_tokens=45,
    strategy="first",
    token_counter=ChatOpenAI(model="gpt-4o"),
)

[SystemMessage(content="you're a good assistant, you always respond with a joke.", additional_kwargs={}, response_metadata={}),
 HumanMessage(content="i wonder why it's called langchain", additional_kwargs={}, response_metadata={})]

编写自定义令牌计数器

我们可以编写一个自定义令牌计数器函数，该函数接收消息列表并返回一个整数。

pip install -qU tiktoken

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

from typing import List

import tiktoken
from langchain_core.messages import BaseMessage, ToolMessage


def str_token_counter(text: str) -> int:
    enc = tiktoken.get_encoding("o200k_base")
    return len(enc.encode(text))


def tiktoken_counter(messages: List[BaseMessage]) -> int:
    """Approximately reproduce https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb

    For simplicity only supports str Message.contents.
    """
    num_tokens = 3  # every reply is primed with <|start|>assistant<|message|>
    tokens_per_message = 3
    tokens_per_name = 1
    for msg in messages:
        if isinstance(msg, HumanMessage):
            role = "user"
        elif isinstance(msg, AIMessage):
            role = "assistant"
        elif isinstance(msg, ToolMessage):
            role = "tool"
        elif isinstance(msg, SystemMessage):
            role = "system"
        else:
            raise ValueError(f"Unsupported messages type {msg.__class__}")
        num_tokens += (
            tokens_per_message
            + str_token_counter(role)
            + str_token_counter(msg.content)
        )
        if msg.name:
            num_tokens += tokens_per_name + str_token_counter(msg.name)
    return num_tokens


trim_messages(
    messages,
    token_counter=tiktoken_counter,
    # Keep the last <= n_count tokens of the messages.
    strategy="last",
    # When token_counter=len, each message
    # will be counted as a single token.
    # Remember to adjust for your use case
    max_tokens=45,
    # Most chat models expect that chat history starts with either:
    # (1) a HumanMessage or
    # (2) a SystemMessage followed by a HumanMessage
    start_on="human",
    # Most chat models expect that chat history ends with either:
    # (1) a HumanMessage or
    # (2) a ToolMessage
    end_on=("human", "tool"),
    # Usually, we want to keep the SystemMessage
    # if it's present in the original history.
    # The SystemMessage has special instructions for the model.
    include_system=True,
)

API 参考:BaseMessage | ToolMessage

[SystemMessage(content="you're a good assistant, you always respond with a joke.", additional_kwargs={}, response_metadata={}),
 HumanMessage(content='what do you call a speechless parrot', additional_kwargs={}, response_metadata={})]

链接

trim_messages 可以像上面一样以命令式方式使用，也可以以声明式方式使用，从而使其易于与链中的其他组件组合使用

llm = ChatOpenAI(model="gpt-4o")

# Notice we don't pass in messages. This creates
# a RunnableLambda that takes messages as input
trimmer = trim_messages(
    token_counter=llm,
    # Keep the last <= n_count tokens of the messages.
    strategy="last",
    # When token_counter=len, each message
    # will be counted as a single token.
    # Remember to adjust for your use case
    max_tokens=45,
    # Most chat models expect that chat history starts with either:
    # (1) a HumanMessage or
    # (2) a SystemMessage followed by a HumanMessage
    start_on="human",
    # Most chat models expect that chat history ends with either:
    # (1) a HumanMessage or
    # (2) a ToolMessage
    end_on=("human", "tool"),
    # Usually, we want to keep the SystemMessage
    # if it's present in the original history.
    # The SystemMessage has special instructions for the model.
    include_system=True,
)

chain = trimmer | llm
chain.invoke(messages)

AIMessage(content='A polygon! Because it\'s a "poly-gone" quiet!', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 13, 'prompt_tokens': 32, 'total_tokens': 45, 'completion_tokens_details': {'reasoning_tokens': 0}}, 'model_name': 'gpt-4o-2024-05-13', 'system_fingerprint': 'fp_057232b607', 'finish_reason': 'stop', 'logprobs': None}, id='run-4fa026e7-9137-4fef-b596-54243615e3b3-0', usage_metadata={'input_tokens': 32, 'output_tokens': 13, 'total_tokens': 45})

查看 LangSmith 跟踪，我们可以看到消息在传递到模型之前首先被修剪： https://smith.langchain.com/public/65af12c4-c24d-4824-90f0-6547566e59bb/r

仅查看修剪器，我们可以看到它是一个 Runnable 对象，可以像所有 Runnables 一样调用

trimmer.invoke(messages)

[SystemMessage(content="you're a good assistant, you always respond with a joke.", additional_kwargs={}, response_metadata={}),
 HumanMessage(content='what do you call a speechless parrot', additional_kwargs={}, response_metadata={})]

与 ChatMessageHistory 一起使用

当处理聊天历史记录时，修剪消息特别有用，聊天历史记录可能会变得任意长

from langchain_core.chat_history import InMemoryChatMessageHistory
from langchain_core.runnables.history import RunnableWithMessageHistory

chat_history = InMemoryChatMessageHistory(messages=messages[:-1])


def dummy_get_session_history(session_id):
    if session_id != "1":
        return InMemoryChatMessageHistory()
    return chat_history


llm = ChatOpenAI(model="gpt-4o")

trimmer = trim_messages(
    max_tokens=45,
    strategy="last",
    token_counter=llm,
    # Usually, we want to keep the SystemMessage
    # if it's present in the original history.
    # The SystemMessage has special instructions for the model.
    include_system=True,
    # Most chat models expect that chat history starts with either:
    # (1) a HumanMessage or
    # (2) a SystemMessage followed by a HumanMessage
    # start_on="human" makes sure we produce a valid chat history
    start_on="human",
)

chain = trimmer | llm
chain_with_history = RunnableWithMessageHistory(chain, dummy_get_session_history)
chain_with_history.invoke(
    [HumanMessage("what do you call a speechless parrot")],
    config={"configurable": {"session_id": "1"}},
)

API 参考:InMemoryChatMessageHistory | RunnableWithMessageHistory

AIMessage(content='A "polygon"!', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 4, 'prompt_tokens': 32, 'total_tokens': 36, 'completion_tokens_details': {'reasoning_tokens': 0}}, 'model_name': 'gpt-4o-2024-05-13', 'system_fingerprint': 'fp_c17d3befe7', 'finish_reason': 'stop', 'logprobs': None}, id='run-71d9fce6-bb0c-4bb3-acc8-d5eaee6ae7bc-0', usage_metadata={'input_tokens': 32, 'output_tokens': 4, 'total_tokens': 36})

查看 LangSmith 跟踪，我们可以看到我们检索了所有消息，但在将消息传递到模型之前，它们被修剪为仅包含系统消息和最后一条人为消息：https://smith.langchain.com/public/17dd700b-9994-44ca-930c-116e00997315/r

API 参考

有关所有参数的完整说明，请访问 API 参考：https://python.langchain.ac.cn/api_reference/core/messages/langchain_core.messages.utils.trim_messages.html