如何使用多模态提示

这里演示了如何使用提示模板来格式化多模态输入以供模型使用。

在这个例子中，我们将要求一个模型描述一张图片。

先决条件

本指南假定您熟悉以下概念

• 聊天模型
• LangChain 工具

npm

npm i axios @langchain/openai @langchain/core

yarn

yarn add axios @langchain/openai @langchain/core

pnpm

pnpm add axios @langchain/openai @langchain/core

import axios from "axios";

const imageUrl =
  "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg";
const axiosRes = await axios.get(imageUrl, { responseType: "arraybuffer" });
const base64 = btoa(
  new Uint8Array(axiosRes.data).reduce(
    (data, byte) => data + String.fromCharCode(byte),
    ""
  )
);

import { ChatPromptTemplate } from "@langchain/core/prompts";
import { ChatOpenAI } from "@langchain/openai";

const model = new ChatOpenAI({ model: "gpt-4o" });

const prompt = ChatPromptTemplate.fromMessages([
  ["system", "Describe the image provided"],
  [
    "user",
    [{ type: "image_url", image_url: "data:image/jpeg;base64,{base64}" }],
  ],
]);

const chain = prompt.pipe(model);

const response = await chain.invoke({ base64 });
console.log(response.content);

The image depicts a scenic outdoor landscape featuring a wooden boardwalk path extending forward through a large field of green grass and vegetation. On either side of the path, the grass is lush and vibrant, with a variety of bushes and low shrubs visible as well. The sky overhead is expansive and mostly clear, adorned with soft, wispy clouds, illuminated by the light giving a warm and serene ambiance. In the distant background, there are clusters of trees and additional foliage, suggesting a natural and tranquil setting, ideal for a peaceful walk or nature exploration.

我们也可以传入多张图片。

const promptWithMultipleImages = ChatPromptTemplate.fromMessages([
  ["system", "compare the two pictures provided"],
  [
    "user",
    [
      {
        type: "image_url",
        image_url: "data:image/jpeg;base64,{imageData1}",
      },
      {
        type: "image_url",
        image_url: "data:image/jpeg;base64,{imageData2}",
      },
    ],
  ],
]);

const chainWithMultipleImages = promptWithMultipleImages.pipe(model);

const res = await chainWithMultipleImages.invoke({
  imageData1: base64,
  imageData2: base64,
});
console.log(res.content);

The two images provided are identical. Both show a wooden boardwalk path extending into a grassy field under a blue sky with scattered clouds. The scenery includes green shrubs and trees in the background, with a bright and clear sky above.