anthropic_tools

危险

由于 Anthropic 现在正式支持工具，因此此 API 已弃用。点击此处阅读文档。

Anthropic 工具

LangChain 提供了一个围绕 Anthropic 的实验性包装器，使其具有与 OpenAI Functions 相同的 API。

设置

首先，安装 @langchain/anthropic 集成包。

提示

查看此部分了解有关安装集成包的常规说明。

npm

npm install @langchain/anthropic

Yarn

yarn add @langchain/anthropic

pnpm

pnpm add @langchain/anthropic

初始化模型

您可以通过与初始化标准 ChatAnthropic 实例相同的方式来初始化此包装器

提示

我们正在统一所有包的模型参数。现在建议使用 model 而不是 modelName，以及 apiKey 用于 API 密钥。

import { ChatAnthropicTools } from "@langchain/anthropic/experimental";

const model = new ChatAnthropicTools({
  temperature: 0.1,
  model: "claude-3-sonnet-20240229",
  apiKey: "YOUR-API-KEY", // In Node.js defaults to process.env.ANTHROPIC_API_KEY
});

传入工具

您现在可以像 OpenAI 一样传入工具

import { ChatAnthropicTools } from "@langchain/anthropic/experimental";
import { HumanMessage } from "@langchain/core/messages";

const model = new ChatAnthropicTools({
  temperature: 0.1,
  model: "claude-3-sonnet-20240229",
}).bind({
  tools: [
    {
      type: "function",
      function: {
        name: "get_current_weather",
        description: "Get the current weather in a given location",
        parameters: {
          type: "object",
          properties: {
            location: {
              type: "string",
              description: "The city and state, e.g. San Francisco, CA",
            },
            unit: { type: "string", enum: ["celsius", "fahrenheit"] },
          },
          required: ["location"],
        },
      },
    },
  ],
  // You can set the `function_call` arg to force the model to use a function
  tool_choice: {
    type: "function",
    function: {
      name: "get_current_weather",
    },
  },
});

const response = await model.invoke([
  new HumanMessage({
    content: "What's the weather in Boston?",
  }),
]);

console.log(response);

/*
  AIMessage {
    lc_serializable: true,
    lc_kwargs: { content: '', additional_kwargs: { tool_calls: [Array] } },
    lc_namespace: [ 'langchain_core', 'messages' ],
    content: '',
    name: undefined,
    additional_kwargs: { tool_calls: [ [Object] ] }
  }
*/

console.log(response.additional_kwargs.tool_calls);

/*
  [
    {
      id: '0',
      type: 'function',
      function: {
        name: 'get_current_weather',
        arguments: '{"location":"Boston, MA","unit":"fahrenheit"}'
      }
    }
  ]
*/

API 参考

• ChatAnthropicTools 来自 @langchain/anthropic/experimental
• HumanMessage 来自 @langchain/core/messages

并行工具调用

模型可以选择调用多个工具。以下是用提取用例的示例

import { z } from "zod";
import { zodToJsonSchema } from "zod-to-json-schema";

import { ChatAnthropicTools } from "@langchain/anthropic/experimental";
import { PromptTemplate } from "@langchain/core/prompts";
import { JsonOutputToolsParser } from "@langchain/core/output_parsers/openai_tools";

const EXTRACTION_TEMPLATE = `Extract and save the relevant entities mentioned in the following passage together with their properties.

Passage:
{input}
`;

const prompt = PromptTemplate.fromTemplate(EXTRACTION_TEMPLATE);

// Use Zod for easier schema declaration
const schema = z.object({
  name: z.string().describe("The name of a person"),
  height: z.number().describe("The person's height"),
  hairColor: z.optional(z.string()).describe("The person's hair color"),
});

const model = new ChatAnthropicTools({
  temperature: 0.1,
  model: "claude-3-sonnet-20240229",
}).bind({
  tools: [
    {
      type: "function",
      function: {
        name: "person",
        description: "Extracts the relevant people from the passage.",
        parameters: zodToJsonSchema(schema),
      },
    },
  ],
  // Can also set to "auto" to let the model choose a tool
  tool_choice: {
    type: "function",
    function: {
      name: "person",
    },
  },
});

// Use a JsonOutputToolsParser to get the parsed JSON response directly.
const chain = await prompt.pipe(model).pipe(new JsonOutputToolsParser());

const response = await chain.invoke({
  input:
    "Alex is 5 feet tall. Claudia is 1 foot taller than Alex and jumps higher than him. Claudia is a brunette and Alex is blonde.",
});

console.log(JSON.stringify(response, null, 2));

/*
  [
    {
      "type": "person",
      "args": {
        "name": "Alex",
        "height": 5,
        "hairColor": "blonde"
      }
    },
    {
      "type": "person",
      "args": {
        "name": "Claudia",
        "height": 6,
        "hairColor": "brunette"
      }
    }
  ]
*/

API 参考

• ChatAnthropicTools 来自 @langchain/anthropic/experimental
• PromptTemplate 来自 @langchain/core/prompts
• JsonOutputToolsParser 来自 @langchain/core/output_parsers/openai_tools

.withStructuredOutput({ ... })

信息

.withStructuredOutput 方法处于测试阶段。它正在积极开发中，因此 API 可能会发生变化。

使用 .withStructuredOutput 方法，您只需一个 Zod 或 JSON 模式，就可以使 LLM 返回结构化输出

import { z } from "zod";
import { ChatAnthropicTools } from "@langchain/anthropic/experimental";
import { ChatPromptTemplate } from "@langchain/core/prompts";

const calculatorSchema = z.object({
  operation: z
    .enum(["add", "subtract", "multiply", "divide"])
    .describe("The type of operation to execute"),
  number1: z.number().describe("The first number to operate on."),
  number2: z.number().describe("The second number to operate on."),
});

const model = new ChatAnthropicTools({
  model: "claude-3-sonnet-20240229",
  temperature: 0.1,
});

// Pass the schema and tool name to the withStructuredOutput method
const modelWithTool = model.withStructuredOutput(calculatorSchema);

// You can also set force: false to allow the model scratchpad space.
// This may improve reasoning capabilities.
// const modelWithTool = model.withStructuredOutput(calculatorSchema, {
//   force: false,
// });

const prompt = ChatPromptTemplate.fromMessages([
  [
    "system",
    "You are a helpful assistant who always needs to use a calculator.",
  ],
  ["human", "{input}"],
]);

// Chain your prompt and model together
const chain = prompt.pipe(modelWithTool);

const response = await chain.invoke({
  input: "What is 2 + 2?",
});
console.log(response);
/*
  { operation: 'add', number1: 2, number2: 2 }
*/

API 参考

• ChatAnthropicTools 来自 @langchain/anthropic/experimental
• ChatPromptTemplate 来自 @langchain/core/prompts

使用 JSON 模式：

import { ChatAnthropicTools } from "@langchain/anthropic/experimental";
import { ChatPromptTemplate } from "@langchain/core/prompts";

const calculatorJsonSchema = {
  type: "object",
  properties: {
    operation: {
      type: "string",
      enum: ["add", "subtract", "multiply", "divide"],
      description: "The type of operation to execute.",
    },
    number1: { type: "number", description: "The first number to operate on." },
    number2: {
      type: "number",
      description: "The second number to operate on.",
    },
  },
  required: ["operation", "number1", "number2"],
  description: "A simple calculator tool",
};

const model = new ChatAnthropicTools({
  model: "claude-3-sonnet-20240229",
  temperature: 0.1,
});

// Pass the schema and optionally, the tool name to the withStructuredOutput method
const modelWithTool = model.withStructuredOutput(calculatorJsonSchema, {
  name: "calculator",
});

const prompt = ChatPromptTemplate.fromMessages([
  [
    "system",
    "You are a helpful assistant who always needs to use a calculator.",
  ],
  ["human", "{input}"],
]);

// Chain your prompt and model together
const chain = prompt.pipe(modelWithTool);

const response = await chain.invoke({
  input: "What is 2 + 2?",
});
console.log(response);
/*
  { operation: 'add', number1: 2, number2: 2 }
*/

API 参考

• ChatAnthropicTools 来自 @langchain/anthropic/experimental
• ChatPromptTemplate 来自 @langchain/core/prompts

相关

• 聊天模型 概念指南
• 聊天模型 操作指南