使用调用模型 API 在 Amazon Bedrock 上调用 Anthropic Claude 模型

异步调用 Anthropic Claude 2 基础模型来生成文本。

        /// <summary>
        /// Asynchronously invokes the Anthropic Claude 2 model to run an inference based on the provided input.
        /// </summary>
        /// <param name="prompt">The prompt that you want Claude to complete.</param>
        /// <returns>The inference response from the model</returns>
        /// <remarks>
        /// 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
        /// </remarks>
        public static async Task<string> InvokeClaudeAsync(string prompt)
        {
            string claudeModelId = "anthropic.claude-v2";

            // Claude requires you to enclose the prompt as follows:
            string enclosedPrompt = "Human: " + prompt + "\n\nAssistant:";

            AmazonBedrockRuntimeClient client = new(RegionEndpoint.USEast1);

            string payload = new JsonObject()
            {
                { "prompt", enclosedPrompt },
                { "max_tokens_to_sample", 200 },
                { "temperature", 0.5 },
                { "stop_sequences", new JsonArray("\n\nHuman:") }
            }.ToJsonString();

            string generatedText = "";
            try
            {
                InvokeModelResponse response = await client.InvokeModelAsync(new InvokeModelRequest()
                {
                    ModelId = claudeModelId,
                    Body = AWSSDKUtils.GenerateMemoryStreamFromString(payload),
                    ContentType = "application/json",
                    Accept = "application/json"
                });

                if (response.HttpStatusCode == System.Net.HttpStatusCode.OK)
                {
                    return JsonNode.ParseAsync(response.Body).Result?["completion"]?.GetValue<string>() ?? "";
                }
                else
                {
                    Console.WriteLine("InvokeModelAsync failed with status code " + response.HttpStatusCode);
                }
            }
            catch (AmazonBedrockRuntimeException e)
            {
                Console.WriteLine(e.Message);
            }
            return generatedText;
        }

调用 Anthropic Claude 2 基础模型以生成文本。

// Each model provider has their own 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

type ClaudeRequest struct {
	Prompt            string   `json:"prompt"`
	MaxTokensToSample int      `json:"max_tokens_to_sample"`
	Temperature       float64  `json:"temperature,omitempty"`
	StopSequences     []string `json:"stop_sequences,omitempty"`
}

type ClaudeResponse struct {
	Completion string `json:"completion"`
}

// Invokes Anthropic Claude on Amazon Bedrock to run an inference using the input
// provided in the request body.
func (wrapper InvokeModelWrapper) InvokeClaude(prompt string) (string, error) {
	modelId := "anthropic.claude-v2"

	// Anthropic Claude requires enclosing the prompt as follows:
	enclosedPrompt := "Human: " + prompt + "\n\nAssistant:"

	body, err := json.Marshal(ClaudeRequest{
		Prompt:            enclosedPrompt,
		MaxTokensToSample: 200,
		Temperature:       0.5,
		StopSequences:     []string{"\n\nHuman:"},
	})

	if err != nil {
		log.Fatal("failed to marshal", err)
	}

	output, err := wrapper.BedrockRuntimeClient.InvokeModel(context.TODO(), &bedrockruntime.InvokeModelInput{
		ModelId:     aws.String(modelId),
		ContentType: aws.String("application/json"),
		Body:        body,
	})

	if err != nil {
		ProcessError(err, modelId)
	}

	var response ClaudeResponse
	if err := json.Unmarshal(output.Body, &response); err != nil {
		log.Fatal("failed to unmarshal", err)
	}

	return response.Completion, nil
}

使用同步客户端调用 Claude 2.x（向下滚动查看异步示例）。

        /**
         * Invokes the Anthropic Claude 2 model to run an inference based on the
         * provided input.
         *
         * @param prompt The prompt for Claude to complete.
         * @return The generated response.
         */
        public static String invokeClaude(String prompt) {
                /*
                 * 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
                 */

                String claudeModelId = "anthropic.claude-v2";

                // Claude requires you to enclose the prompt as follows:
                String enclosedPrompt = "Human: " + prompt + "\n\nAssistant:";

                BedrockRuntimeClient client = BedrockRuntimeClient.builder()
                                .region(Region.US_EAST_1)
                                .credentialsProvider(ProfileCredentialsProvider.create())
                                .build();

                String payload = new JSONObject()
                                .put("prompt", enclosedPrompt)
                                .put("max_tokens_to_sample", 200)
                                .put("temperature", 0.5)
                                .put("stop_sequences", List.of("\n\nHuman:"))
                                .toString();

                InvokeModelRequest request = InvokeModelRequest.builder()
                                .body(SdkBytes.fromUtf8String(payload))
                                .modelId(claudeModelId)
                                .contentType("application/json")
                                .accept("application/json")
                                .build();

                InvokeModelResponse response = client.invokeModel(request);

                JSONObject responseBody = new JSONObject(response.body().asUtf8String());

                String generatedText = responseBody.getString("completion");

                return generatedText;
        }

使用异步客户端调用 Claude 2.x。

    /**
     * Asynchronously invokes the Anthropic Claude 2 model to run an inference based
     * on the provided input.
     *
     * @param prompt The prompt that you want Claude to complete.
     * @return The inference response from the model.
     */
    public static String invokeClaude(String prompt) {
        /*
         * 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
         */

        String claudeModelId = "anthropic.claude-v2";

        // Claude requires you to enclose the prompt as follows:
        String enclosedPrompt = "Human: " + prompt + "\n\nAssistant:";

        BedrockRuntimeAsyncClient client = BedrockRuntimeAsyncClient.builder()
                .region(Region.US_EAST_1)
                .credentialsProvider(ProfileCredentialsProvider.create())
                .build();

        String payload = new JSONObject()
                .put("prompt", enclosedPrompt)
                .put("max_tokens_to_sample", 200)
                .put("temperature", 0.5)
                .put("stop_sequences", List.of("\n\nHuman:"))
                .toString();

        InvokeModelRequest request = InvokeModelRequest.builder()
                .body(SdkBytes.fromUtf8String(payload))
                .modelId(claudeModelId)
                .contentType("application/json")
                .accept("application/json")
                .build();

        CompletableFuture<InvokeModelResponse> completableFuture = client.invokeModel(request)
                .whenComplete((response, exception) -> {
                    if (exception != null) {
                        System.out.println("Model invocation failed: " + exception);
                    }
                });

        String generatedText = "";
        try {
            InvokeModelResponse response = completableFuture.get();
            JSONObject responseBody = new JSONObject(response.body().asUtf8String());
            generatedText = responseBody.getString("completion");
        } catch (InterruptedException e) {
            Thread.currentThread().interrupt();
            System.err.println(e.getMessage());
        } catch (ExecutionException e) {
            System.err.println(e.getMessage());
        }

        return generatedText;
    }

使用调用模型 API 发送短信。

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

import { fileURLToPath } from "url";

import { FoundationModels } from "../../config/foundation_models.js";
import {
  BedrockRuntimeClient,
  InvokeModelCommand,
  InvokeModelWithResponseStreamCommand,
} from "@aws-sdk/client-bedrock-runtime";

/**
 * @typedef {Object} ResponseContent
 * @property {string} text
 *
 * @typedef {Object} MessagesResponseBody
 * @property {ResponseContent[]} content
 *
 * @typedef {Object} Delta
 * @property {string} text
 *
 * @typedef {Object} Message
 * @property {string} role
 *
 * @typedef {Object} Chunk
 * @property {string} type
 * @property {Delta} delta
 * @property {Message} message
 */

/**
 * Invokes Anthropic Claude 3 using the Messages API.
 *
 * To learn more about the Anthropic Messages API, go to:
 * https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-anthropic-claude-messages.html
 *
 * @param {string} prompt - The input text prompt for the model to complete.
 * @param {string} [modelId] - The ID of the model to use. Defaults to "anthropic.claude-3-haiku-20240307-v1:0".
 */
export const invokeModel = async (
  prompt,
  modelId = "anthropic.claude-3-haiku-20240307-v1:0",
) => {
  // Create a new Bedrock Runtime client instance.
  const client = new BedrockRuntimeClient({ region: "us-east-1" });

  // Prepare the payload for the model.
  const payload = {
    anthropic_version: "bedrock-2023-05-31",
    max_tokens: 1000,
    messages: [
      {
        role: "user",
        content: [{ type: "text", text: prompt }],
      },
    ],
  };

  // Invoke Claude with the payload and wait for the response.
  const command = new InvokeModelCommand({
    contentType: "application/json",
    body: JSON.stringify(payload),
    modelId,
  });
  const apiResponse = await client.send(command);

  // Decode and return the response(s)
  const decodedResponseBody = new TextDecoder().decode(apiResponse.body);
  /** @type {MessagesResponseBody} */
  const responseBody = JSON.parse(decodedResponseBody);
  return responseBody.content[0].text;
};

/**
 * Invokes Anthropic Claude 3 and processes the response stream.
 *
 * To learn more about the Anthropic Messages API, go to:
 * https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-anthropic-claude-messages.html
 *
 * @param {string} prompt - The input text prompt for the model to complete.
 * @param {string} [modelId] - The ID of the model to use. Defaults to "anthropic.claude-3-haiku-20240307-v1:0".
 */
export const invokeModelWithResponseStream = async (
  prompt,
  modelId = "anthropic.claude-3-haiku-20240307-v1:0",
) => {
  // Create a new Bedrock Runtime client instance.
  const client = new BedrockRuntimeClient({ region: "us-east-1" });

  // Prepare the payload for the model.
  const payload = {
    anthropic_version: "bedrock-2023-05-31",
    max_tokens: 1000,
    messages: [
      {
        role: "user",
        content: [{ type: "text", text: prompt }],
      },
    ],
  };

  // Invoke Claude with the payload and wait for the API to respond.
  const command = new InvokeModelWithResponseStreamCommand({
    contentType: "application/json",
    body: JSON.stringify(payload),
    modelId,
  });
  const apiResponse = await client.send(command);

  let completeMessage = "";

  // Decode and process the response stream
  for await (const item of apiResponse.body) {
    /** @type Chunk */
    const chunk = JSON.parse(new TextDecoder().decode(item.chunk.bytes));
    const chunk_type = chunk.type;

    if (chunk_type === "content_block_delta") {
      const text = chunk.delta.text;
      completeMessage = completeMessage + text;
      process.stdout.write(text);
    }
  }

  // Return the final response
  return completeMessage;
};

// Invoke the function if this file was run directly.
if (process.argv[1] === fileURLToPath(import.meta.url)) {
  const prompt = 'Write a paragraph starting with: "Once upon a time..."';
  const modelId = FoundationModels.CLAUDE_3_HAIKU.modelId;
  console.log(`Prompt: ${prompt}`);
  console.log(`Model ID: ${modelId}`);

  try {
    console.log("-".repeat(53));
    const response = await invokeModel(prompt, modelId);
    console.log("\n" + "-".repeat(53));
    console.log("Final structured response:");
    console.log(response);
  } catch (err) {
    console.log(`\n${err}`);
  }
}

调用 Anthropic Claude 2 基础模型以生成文本。

    public function invokeClaude($prompt)
    {
        # 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

        $completion = "";

        try {
            $modelId = 'anthropic.claude-v2';

            # Claude requires you to enclose the prompt as follows:
            $prompt = "\n\nHuman: {$prompt}\n\nAssistant:";

            $body = [
                'prompt' => $prompt,
                'max_tokens_to_sample' => 200,
                'temperature' => 0.5,
                'stop_sequences' => ["\n\nHuman:"],
            ];

            $result = $this->bedrockRuntimeClient->invokeModel([
                'contentType' => 'application/json',
                'body' => json_encode($body),
                'modelId' => $modelId,
            ]);

            $response_body = json_decode($result['body']);

            $completion = $response_body->completion;
        } catch (Exception $e) {
            echo "Error: ({$e->getCode()}) - {$e->getMessage()}\n";
        }

        return $completion;
    }

使用调用模型 API 发送短信。

# Use the native inference API to send a text message to Anthropic Claude.

import boto3
import json

# Create a Bedrock Runtime client in the AWS Region of your choice.
client = boto3.client("bedrock-runtime", region_name="us-east-1")

# Set the model ID, e.g., Claude 3 Haiku.
model_id = "anthropic.claude-3-haiku-20240307-v1:0"

# Define the prompt for the model.
prompt = "Describe the purpose of a 'hello world' program in one line."

# Format the request payload using the model's native structure.
native_request = {
    "anthropic_version": "bedrock-2023-05-31",
    "max_tokens": 512,
    "temperature": 0.5,
    "messages": [
        {
            "role": "user",
            "content": [{"type": "text", "text": prompt}],
        }
    ],
}

# Convert the native request to JSON.
request = json.dumps(native_request)

# Invoke the model with the request.
response = client.invoke_model(modelId=model_id, body=request)

# Decode the response body.
model_response = json.loads(response["body"].read())

# Extract and print the response text.
response_text = model_response["content"][0]["text"]
print(response_text)

调用 Anthropic Claude 2 基础模型以生成文本。此示例使用了 /US2/CL_JSON 的功能，这些功能在某些版本上可能不可用。 NetWeaver

    "Claude V2 Input Parameters should be in a format like this:
*   {
*     "prompt":"\n\nHuman:\\nTell me a joke\n\nAssistant:\n",
*     "max_tokens_to_sample":2048,
*     "temperature":0.5,
*     "top_k":250,
*     "top_p":1.0,
*     "stop_sequences":[]
*   }

    DATA: BEGIN OF ls_input,
            prompt               TYPE string,
            max_tokens_to_sample TYPE /aws1/rt_shape_integer,
            temperature          TYPE /aws1/rt_shape_float,
            top_k                TYPE /aws1/rt_shape_integer,
            top_p                TYPE /aws1/rt_shape_float,
            stop_sequences       TYPE /aws1/rt_stringtab,
          END OF ls_input.

    "Leave ls_input-stop_sequences empty.
    ls_input-prompt = |\n\nHuman:\\n{ iv_prompt }\n\nAssistant:\n|.
    ls_input-max_tokens_to_sample = 2048.
    ls_input-temperature = '0.5'.
    ls_input-top_k = 250.
    ls_input-top_p = 1.

    "Serialize into JSON with /ui2/cl_json -- this assumes SAP_UI is installed.
    DATA(lv_json) = /ui2/cl_json=>serialize(
      data = ls_input
                pretty_name   = /ui2/cl_json=>pretty_mode-low_case ).

    TRY.
        DATA(lo_response) = lo_bdr->invokemodel(
          iv_body = /aws1/cl_rt_util=>string_to_xstring( lv_json )
          iv_modelid = 'anthropic.claude-v2'
          iv_accept = 'application/json'
          iv_contenttype = 'application/json' ).

        "Claude V2 Response format will be:
*       {
*         "completion": "Knock Knock...",
*         "stop_reason": "stop_sequence"
*       }
        DATA: BEGIN OF ls_response,
                completion  TYPE string,
                stop_reason TYPE string,
              END OF ls_response.

        /ui2/cl_json=>deserialize(
          EXPORTING jsonx = lo_response->get_body( )
                    pretty_name = /ui2/cl_json=>pretty_mode-camel_case
          CHANGING  data  = ls_response ).

        DATA(lv_answer) = ls_response-completion.
      CATCH /aws1/cx_bdraccessdeniedex INTO DATA(lo_ex).
        WRITE / lo_ex->get_text( ).
        WRITE / |Don't forget to enable model access at https://console.aws.amazon.com/bedrock/home?#/modelaccess|.

    ENDTRY.

调用 Anthropic Claude 2 基础模型，使用 L2 高级客户端生成文本。

    TRY.
        DATA(lo_bdr_l2_claude) = /aws1/cl_bdr_l2_factory=>create_claude_2( lo_bdr ).
        " iv_prompt can contain a prompt like 'tell me a joke about Java programmers'.
        DATA(lv_answer) = lo_bdr_l2_claude->prompt_for_text( iv_prompt ).
      CATCH /aws1/cx_bdraccessdeniedex INTO DATA(lo_ex).
        WRITE / lo_ex->get_text( ).
        WRITE / |Don't forget to enable model access at https://console.aws.amazon.com/bedrock/home?#/modelaccess|.

    ENDTRY.