如何合并相同类型的连续消息

mergeMessageRuns 函数可在 @langchain/core 版本 0.2.8 及以上版本中使用。

某些模型不支持连续传递相同类型的（连续的）消息。

mergeMessageRuns 工具可以轻松合并相同类型的连续消息。

基本用法

import {
  HumanMessage,
  SystemMessage,
  AIMessage,
  mergeMessageRuns,
} from "@langchain/core/messages";

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

const merged = mergeMessageRuns(messages);
console.log(
  merged
    .map((x) =>
      JSON.stringify(
        {
          role: x._getType(),
          content: x.content,
        },
        null,
        2
      )
    )
    .join("\n\n")
);

{
  "role": "system",
  "content": "you're a good assistant.\nyou always respond with a joke."
}

{
  "role": "human",
  "content": [
    {
      "type": "text",
      "text": "i wonder why it's called langchain"
    },
    {
      "type": "text",
      "text": "and who is harrison chasing anyways"
    }
  ]
}

{
  "role": "ai",
  "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!"
}

请注意，如果要合并的消息中的内容是内容块的列表，则合并后的消息将具有内容块的列表。如果要合并的两条消息的内容都是字符串，则它们的内容会用换行符连接起来。

链式操作

mergeMessageRuns 可以以命令式（如上所示）或声明式的方式使用，便于与链中的其他组件组合使用：

import { ChatAnthropic } from "@langchain/anthropic";
import { mergeMessageRuns } from "@langchain/core/messages";

const llm = new ChatAnthropic({
  model: "claude-3-sonnet-20240229",
  temperature: 0,
});
// Notice we don't pass in messages. This creates
// a RunnableLambda that takes messages as input
const merger = mergeMessageRuns();
const chain = merger.pipe(llm);
await chain.invoke(messages);

AIMessage {
  lc_serializable: true,
  lc_kwargs: {
    content: [],
    additional_kwargs: {
      id: 'msg_01LsdS4bjQ3EznH7Tj4xujV1',
      type: 'message',
      role: 'assistant',
      model: 'claude-3-sonnet-20240229',
      stop_reason: 'end_turn',
      stop_sequence: null,
      usage: [Object]
    },
    tool_calls: [],
    usage_metadata: { input_tokens: 84, output_tokens: 3, total_tokens: 87 },
    invalid_tool_calls: [],
    response_metadata: {}
  },
  lc_namespace: [ 'langchain_core', 'messages' ],
  content: [],
  name: undefined,
  additional_kwargs: {
    id: 'msg_01LsdS4bjQ3EznH7Tj4xujV1',
    type: 'message',
    role: 'assistant',
    model: 'claude-3-sonnet-20240229',
    stop_reason: 'end_turn',
    stop_sequence: null,
    usage: { input_tokens: 84, output_tokens: 3 }
  },
  response_metadata: {
    id: 'msg_01LsdS4bjQ3EznH7Tj4xujV1',
    model: 'claude-3-sonnet-20240229',
    stop_reason: 'end_turn',
    stop_sequence: null,
    usage: { input_tokens: 84, output_tokens: 3 }
  },
  id: undefined,
  tool_calls: [],
  invalid_tool_calls: [],
  usage_metadata: { input_tokens: 84, output_tokens: 3, total_tokens: 87 }
}

查看 LangSmith 跟踪，我们可以看到消息在传递给模型之前会被合并。

仅查看合并过程，我们可以看到它是一个 Runnable 对象，可以像所有 Runnable 一样被调用：

await merger.invoke(messages);

[
  SystemMessage {
    lc_serializable: true,
    lc_kwargs: {
      content: "you're a good assistant.\nyou always respond with a joke.",
      name: undefined,
      additional_kwargs: {},
      response_metadata: {},
      id: undefined
    },
    lc_namespace: [ 'langchain_core', 'messages' ],
    content: "you're a good assistant.\nyou always respond with a joke.",
    name: undefined,
    additional_kwargs: {},
    response_metadata: {},
    id: undefined
  },
  HumanMessage {
    lc_serializable: true,
    lc_kwargs: {
      content: [Array],
      name: undefined,
      additional_kwargs: {},
      response_metadata: {},
      id: undefined
    },
    lc_namespace: [ 'langchain_core', 'messages' ],
    content: [ [Object], [Object] ],
    name: undefined,
    additional_kwargs: {},
    response_metadata: {},
    id: undefined
  },
  AIMessage {
    lc_serializable: true,
    lc_kwargs: {
      content: `Well, I guess they thought "WordRope" and "SentenceString" just didn't have the same ring to it!\n` +
        "Why, he's probably chasing after the last cup of coffee in the office!",
      name: undefined,
      additional_kwargs: {},
      response_metadata: {},
      id: undefined,
      tool_calls: [],
      invalid_tool_calls: [],
      usage_metadata: undefined
    },
    lc_namespace: [ 'langchain_core', 'messages' ],
    content: `Well, I guess they thought "WordRope" and "SentenceString" just didn't have the same ring to it!\n` +
      "Why, he's probably chasing after the last cup of coffee in the office!",
    name: undefined,
    additional_kwargs: {},
    response_metadata: {},
    id: undefined,
    tool_calls: [],
    invalid_tool_calls: [],
    usage_metadata: undefined
  }
]

API 参考

有关所有参数的完整描述，请前往 API 参考。