LangChain v0.0.146中ChatPromptTemplate无法调用问题求助

核心原因 #

LangChain v0.0.146版本中，ChatPromptTemplate和MessagesPlaceholder并不在langchain/prompts的顶级导出里，而是属于聊天专用的子模块，当前导入路径错误，导致无法正确获取构造函数，调用fromMessages时才会提示“不是函数”。

修正步骤 #

1. 调整导入路径 #

将原导入代码：

const { ChatPromptTemplate, MessagesPlaceholder } = require('langchain/prompts');

替换为从聊天子模块导入：

const { ChatPromptTemplate, MessagesPlaceholder } = require('langchain/prompts/chat');

2. 排查其他潜在问题 #

• 确保chatPromptMultipleAdvisors变量已定义，避免运行时出现未定义错误
• BufferMemory中的llm参数在v0.0.146版本中并非必填项，若无需相关功能可考虑移除

修正后的完整代码 #

const { ChatOpenAI } = require("langchain/chat_models/openai");
const { ConversationChain } = require('langchain/chains');
const { BufferMemory } = require('langchain/memory');
// 修正导入路径
const { ChatPromptTemplate, MessagesPlaceholder } = require('langchain/prompts/chat');

// 定义系统提示词变量
const chatPromptMultipleAdvisors = "你的系统提示词内容";

async function runChat() {
    memory1 = new BufferMemory({
        returnMessages: true,
        memoryKey: "memKey1",
        llm: new ChatOpenAI({ modelName: "gpt-3.5-turbo", temperature: 0,
        callbacks: [
            {
              handleLLMEnd: (output, runId, parentRunId, tags) => {
                const { completionTokens, promptTokens, totalTokens } = output.llmOutput?.tokenUsage;
                console.log('Tokens used: ', completionTokens, promptTokens, totalTokens);
              },
            },
          ], }),
      });

      model = new ChatOpenAI();
      
      const prompt1 = ChatPromptTemplate.fromMessages([
        [
          "system",
          chatPromptMultipleAdvisors,
        ],
        new MessagesPlaceholder("memKey1"),
        ["human", "{input}"],
      ]);

      chain1 = new ConversationChain({ llm: model, prompt: prompt1, memory: memory1 });
      res11 = await chain1.call({ input: "what is AI training?" });
      console.log({ res11, memory1: await memory1.loadMemoryVariables({}) });
}

内容的提问来源于stack exchange，提问作者Mor Sagmon