使用适用于 JavaScript (v3) 的 SDK 的亚马逊 Bedrock 运行时示例

调用 AI21 Labs Jurassic-2 基础模型以生成文本。

// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0

import { fileURLToPath } from "url";

import {
  AccessDeniedException,
  BedrockRuntimeClient,
  InvokeModelCommand,
} from "@aws-sdk/client-bedrock-runtime";

/**
 * @typedef {Object} Data
 * @property {string} text
 *
 * @typedef {Object} Completion
 * @property {Data} data
 *
 * @typedef {Object} ResponseBody
 * @property {Completion[]} completions
 */

/**
 * Invokes the AI21 Labs Jurassic-2 large-language model to run an inference
 * using the input provided in the request body.
 *
 * @param {string} prompt - The prompt that you want Jurassic-2 to complete.
 * @returns {string} The inference response (completion) from the model.
 */
export const invokeJurassic2 = async (prompt) => {
  const client = new BedrockRuntimeClient({ region: "us-east-1" });

  const modelId = "ai21.j2-mid-v1";

  /* The different model providers have individual request and response formats.
   * For the format, ranges, and default values for AI21 Labs Jurassic-2, refer to:
   * https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-jurassic2.html
   */
  const payload = {
    prompt,
    maxTokens: 500,
    temperature: 0.5,
  };

  const command = new InvokeModelCommand({
    body: JSON.stringify(payload),
    contentType: "application/json",
    accept: "application/json",
    modelId,
  });

  try {
    const response = await client.send(command);
    const decodedResponseBody = new TextDecoder().decode(response.body);

    /** @type {ResponseBody} */
    const responseBody = JSON.parse(decodedResponseBody);

    return responseBody.completions[0].data.text;
  } catch (err) {
    if (err instanceof AccessDeniedException) {
      console.error(
        `Access denied. Ensure you have the correct permissions to invoke ${modelId}.`,
      );
    } else {
      throw err;
    }
  }
};

// Invoke the function if this file was run directly.
if (process.argv[1] === fileURLToPath(import.meta.url)) {
  const prompt = 'Complete the following: "Once upon a time..."';
  console.log("\nModel: AI21 Labs Jurassic-2");
  console.log(`Prompt: ${prompt}`);

  const completion = await invokeJurassic2(prompt);
  console.log("Completion:");
  console.log(completion);
  console.log("\n");
}

调用 Amazon Titan Text G1 基础模型来生成文本。

// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0

import { fileURLToPath } from "url";

import {
  AccessDeniedException,
  BedrockRuntimeClient,
  InvokeModelCommand,
} from "@aws-sdk/client-bedrock-runtime";

/**
 * @typedef {Object} ResponseBody
 * @property {Object[]} results
 */

/**
 * Invokes the Titan Text G1 - Express model to run an inference
 * using the input provided in the request body.
 *
 * @param {string} prompt - The prompt that you want Titan Text Express to complete.
 * @returns {object[]} The inference response (results) from the model.
 */
export const invokeTitanTextExpressV1 = async (prompt) => {
  const client = new BedrockRuntimeClient({ region: "us-east-1" });

  const modelId = "amazon.titan-text-express-v1";

  /* The different model providers have individual request and response formats.
   * For the format, ranges, and default values for Titan text, refer to:
   * https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-titan-text.html
   */
  const textGenerationConfig = {
    maxTokenCount: 4096,
    stopSequences: [],
    temperature: 0,
    topP: 1,
  };

  const payload = {
    inputText: prompt,
    textGenerationConfig,
  };

  const command = new InvokeModelCommand({
    body: JSON.stringify(payload),
    contentType: "application/json",
    accept: "application/json",
    modelId,
  });

  try {
    const response = await client.send(command);
    const decodedResponseBody = new TextDecoder().decode(response.body);

    /** @type {ResponseBody} */
    const responseBody = JSON.parse(decodedResponseBody);
    return responseBody.results;
  } catch (err) {
    if (err instanceof AccessDeniedException) {
      console.error(
        `Access denied. Ensure you have the correct permissions to invoke ${modelId}.`,
      );
    } else {
      throw err;
    }
  }
};

// Invoke the function if this file was run directly.
if (process.argv[1] === fileURLToPath(import.meta.url)) {
  const prompt = `Meeting transcript: Miguel: Hi Brant, I want to discuss the workstream  
    for our new product launch Brant: Sure Miguel, is there anything in particular you want
    to discuss? Miguel: Yes, I want to talk about how users enter into the product.
    Brant: Ok, in that case let me add in Namita. Namita: Hey everyone 
    Brant: Hi Namita, Miguel wants to discuss how users enter into the product.
    Miguel: its too complicated and we should remove friction.  
    for example, why do I need to fill out additional forms?  
    I also find it difficult to find where to access the product
    when I first land on the landing page. Brant: I would also add that
    I think there are too many steps. Namita: Ok, I can work on the
    landing page to make the product more discoverable but brant
    can you work on the additonal forms? Brant: Yes but I would need 
    to work with James from another team as he needs to unblock the sign up workflow.
    Miguel can you document any other concerns so that I can discuss with James only once?
    Miguel: Sure.
    From the meeting transcript above, Create a list of action items for each person.`;

  console.log("\nModel: Titan Text Express v1");
  console.log(`Prompt: ${prompt}`);

  const results = await invokeTitanTextExpressV1(prompt);
  console.log("Completion:");
  for (const result of results) {
    console.log(result.outputText);
  }
  console.log("\n");
}

调用 Anthropic Claude 2 基础模型以生成文本。

// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0

import { fileURLToPath } from "url";

import {
  AccessDeniedException,
  BedrockRuntimeClient,
  InvokeModelCommand,
} from "@aws-sdk/client-bedrock-runtime";

/**
 * @typedef {Object} ResponseBody
 * @property {string} completion
 */

/**
 * Invokes the Anthropic Claude 2 model to run an inference using the input
 * provided in the request body.
 *
 * @param {string} prompt - The prompt that you want Claude to complete.
 * @returns {string} The inference response (completion) from the model.
 */
export const invokeClaude = async (prompt) => {
  const client = new BedrockRuntimeClient({ region: "us-east-1" });

  const modelId = "anthropic.claude-v2";

  /* Claude requires you to enclose the prompt as follows: */
  const enclosedPrompt = `Human: ${prompt}\n\nAssistant:`;

  /* The different model providers have individual request and response formats.
   * For the format, ranges, and default values for Anthropic Claude, refer to:
   * https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-claude.html
   */
  const payload = {
    prompt: enclosedPrompt,
    max_tokens_to_sample: 500,
    temperature: 0.5,
    stop_sequences: ["\n\nHuman:"],
  };

  const command = new InvokeModelCommand({
    body: JSON.stringify(payload),
    contentType: "application/json",
    accept: "application/json",
    modelId,
  });

  try {
    const response = await client.send(command);
    const decodedResponseBody = new TextDecoder().decode(response.body);

    /** @type {ResponseBody} */
    const responseBody = JSON.parse(decodedResponseBody);

    return responseBody.completion;
  } catch (err) {
    if (err instanceof AccessDeniedException) {
      console.error(
        `Access denied. Ensure you have the correct permissions to invoke ${modelId}.`,
      );
    } else {
      throw err;
    }
  }
};

// Invoke the function if this file was run directly.
if (process.argv[1] === fileURLToPath(import.meta.url)) {
  const prompt = 'Complete the following: "Once upon a time..."';
  console.log("\nModel: Anthropic Claude v2");
  console.log(`Prompt: ${prompt}`);

  const completion = await invokeClaude(prompt);
  console.log("Completion:");
  console.log(completion);
  console.log("\n");
}

调用 Meta Llama 2 Chat 基础模型以生成文本。

// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0

import { fileURLToPath } from "url";

import {
  AccessDeniedException,
  BedrockRuntimeClient,
  InvokeModelCommand,
} from "@aws-sdk/client-bedrock-runtime";

/**
 * @typedef {Object} ResponseBody
 * @property {generation} text
 */

/**
 * Invokes the Meta Llama 2 Chat model to run an inference
 * using the input provided in the request body.
 *
 * @param {string} prompt - The prompt that you want Llama-2 to complete.
 * @returns {string} The inference response (generation) from the model.
 */
export const invokeLlama2 = async (prompt) => {
  const client = new BedrockRuntimeClient({ region: "us-east-1" });

  const modelId = "meta.llama2-13b-chat-v1";

  /* The different model providers have individual request and response formats.
   * For the format, ranges, and default values for Meta Llama 2 Chat, refer to:
   * https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-meta.html
   */
  const payload = {
    prompt,
    temperature: 0.5,
    top_p: 0.9,
    max_gen_len: 512,
  };

  const command = new InvokeModelCommand({
    body: JSON.stringify(payload),
    contentType: "application/json",
    accept: "application/json",
    modelId,
  });

  try {
    const response = await client.send(command);
    const decodedResponseBody = new TextDecoder().decode(response.body);

    /** @type {ResponseBody} */
    const responseBody = JSON.parse(decodedResponseBody);

    return responseBody.generation;
  } catch (err) {
    if (err instanceof AccessDeniedException) {
      console.error(
        `Access denied. Ensure you have the correct permissions to invoke ${modelId}.`,
      );
    } else {
      throw err;
    }
  }
};

// Invoke the function if this file was run directly.
if (process.argv[1] === fileURLToPath(import.meta.url)) {
  const prompt = 'Complete the following: "Once upon a time..."';
  console.log("\nModel: Meta Llama 2 Chat");
  console.log(`Prompt: ${prompt}`);

  const completion = await invokeLlama2(prompt);
  console.log("Completion:");
  console.log(completion);
  console.log("\n");
}

调用 Mistral 7B 基础模型生成文本。

// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0

import { fileURLToPath } from "url";

import {
  AccessDeniedException,
  BedrockRuntimeClient,
  InvokeModelCommand,
} from "@aws-sdk/client-bedrock-runtime";

/**
 * @typedef {Object} Output
 * @property {string} text
 *
 * @typedef {Object} ResponseBody
 * @property {Output[]} outputs
 */

/**
 * Invokes the Mistral 7B model to run an inference using the input
 * provided in the request body.
 *
 * @param {string} prompt - The prompt that you want Mistral to complete.
 * @returns {string[]} A list of inference responses (completions) from the model.
 */
export const invokeMistral7B = async (prompt) => {
  const client = new BedrockRuntimeClient({ region: "us-west-2" });

  const modelId = "mistral.mistral-7b-instruct-v0:2";

  // Mistral instruct models provide optimal results when embedding
  // the prompt into the following template:
  const instruction = `<s>[INST] ${prompt} [/INST]`;

  const payload = {
    prompt: instruction,
    max_tokens: 500,
    temperature: 0.5,
  };

  const command = new InvokeModelCommand({
    body: JSON.stringify(payload),
    contentType: "application/json",
    accept: "application/json",
    modelId,
  });

  try {
    const response = await client.send(command);
    const decodedResponseBody = new TextDecoder().decode(response.body);

    /** @type {ResponseBody} */
    const responseBody = JSON.parse(decodedResponseBody);

    return responseBody.outputs.map((output) => output.text);
  } catch (err) {
    if (err instanceof AccessDeniedException) {
      console.error(
        `Access denied. Ensure you have the correct permissions to invoke ${modelId}.`,
      );
    } else {
      throw err;
    }
  }
};

// Invoke the function if this file was run directly.
if (process.argv[1] === fileURLToPath(import.meta.url)) {
  const prompt = 'Complete the following: "Once upon a time..."';
  console.log("\nModel: Mistral 7B");
  console.log(`Prompt: ${prompt}`);

  const completions = await invokeMistral7B(prompt);
  completions.forEach((completion) => {
    console.log("Completion:");
    console.log(completion);
    console.log("\n");
  });
}

调用 Mixtral 8x7b 基础模型来生成文本。

// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0

import { fileURLToPath } from "url";

import {
  AccessDeniedException,
  BedrockRuntimeClient,
  InvokeModelCommand,
} from "@aws-sdk/client-bedrock-runtime";
import { invokeMistral7B } from "./invoke-mistral7b.js";

/**
 * @typedef {Object} Output
 * @property {string} text
 *
 * @typedef {Object} ResponseBody
 * @property {Output[]} outputs
 */

/**
 * Invokes the Mixtral 8x7B model to run an inference using the input
 * provided in the request body.
 *
 * @param {string} prompt - The prompt that you want Mistral to complete.
 * @returns {string[]} A list of inference responses (completions) from the model.
 */
export const invokeMixtral8x7B = async (prompt) => {
  const client = new BedrockRuntimeClient({ region: "us-west-2" });

  // Mistral instruct models provide optimal results when embedding
  // the prompt into the following template:
  const instruction = `<s>[INST] ${prompt} [/INST]`;

  const modelId = "mistral.mixtral-8x7b-instruct-v0:1";

  const payload = {
    prompt: instruction,
    max_tokens: 500,
    temperature: 0.5,
  };

  const command = new InvokeModelCommand({
    body: JSON.stringify(payload),
    contentType: "application/json",
    accept: "application/json",
    modelId,
  });

  try {
    const response = await client.send(command);
    const decodedResponseBody = new TextDecoder().decode(response.body);

    /** @type {ResponseBody} */
    const responseBody = JSON.parse(decodedResponseBody);

    return responseBody.outputs.map((output) => output.text);
  } catch (err) {
    if (err instanceof AccessDeniedException) {
      console.error(
        `Access denied. Ensure you have the correct permissions to invoke ${modelId}.`,
      );
    } else {
      throw err;
    }
  }
};

// Invoke the function if this file was run directly.
if (process.argv[1] === fileURLToPath(import.meta.url)) {
  const prompt = 'Complete the following: "Once upon a time..."';
  console.log("\nModel: Mixtral 8x7B");
  console.log(`Prompt: ${prompt}`);

  const completions = await invokeMistral7B(prompt);
  completions.forEach((completion) => {
    console.log("Completion:");
    console.log(completion);
    console.log("\n");
  });
}