使用调用模型 API 在亚马逊 Bedrock 上调用 Amazon Titan 文本模型

使用调用模型 API 发送短信。

        /// <summary>
        /// Asynchronously invokes the Amazon Titan Text G1 Express model to run an inference based on the provided input.
        /// </summary>
        /// <param name="prompt">The prompt that you want Amazon Titan Text G1 Express 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 Amazon Titan Text G1 Express, refer to:
        ///     https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-titan-text.html
        /// </remarks>
        public static async Task<string> InvokeTitanTextG1Async(string prompt)
        {
            string titanTextG1ModelId = "amazon.titan-text-express-v1";

            AmazonBedrockRuntimeClient client = new(RegionEndpoint.USEast1);

            string payload = new JsonObject()
            {
                { "inputText", prompt },
                { "textGenerationConfig", new JsonObject()
                    {
                        { "maxTokenCount", 512 },
                        { "temperature", 0f },
                        { "topP", 1f }
                    }
                }
            }.ToJsonString();

            string generatedText = "";
            try
            {
                InvokeModelResponse response = await client.InvokeModelAsync(new InvokeModelRequest()
                {
                    ModelId = titanTextG1ModelId,
                    Body = AWSSDKUtils.GenerateMemoryStreamFromString(payload),
                    ContentType = "application/json",
                    Accept = "application/json"
                });

                if (response.HttpStatusCode == System.Net.HttpStatusCode.OK)
                {
                    var results = JsonNode.ParseAsync(response.Body).Result?["results"]?.AsArray();

                    return results is null ? "" : string.Join(" ", results.Select(x => x?["outputText"]?.GetValue<string?>()));
                }
                else
                {
                    Console.WriteLine("InvokeModelAsync failed with status code " + response.HttpStatusCode);
                }
            }
            catch (AmazonBedrockRuntimeException e)
            {
                Console.WriteLine(e.Message);
            }
            return generatedText;
        }

使用调用模型 API 发送短信。

// Each model provider has their own individual request and response formats.
// For the format, ranges, and default values for Amazon Titan Text, refer to:
// https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-titan-text.html
type TitanTextRequest struct {
	InputText            string               `json:"inputText"`
	TextGenerationConfig TextGenerationConfig `json:"textGenerationConfig"`
}

type TextGenerationConfig struct {
	Temperature   float64  `json:"temperature"`
	TopP          float64  `json:"topP"`
	MaxTokenCount int      `json:"maxTokenCount"`
	StopSequences []string `json:"stopSequences,omitempty"`
}

type TitanTextResponse struct {
	InputTextTokenCount int      `json:"inputTextTokenCount"`
	Results             []Result `json:"results"`
}

type Result struct {
	TokenCount       int    `json:"tokenCount"`
	OutputText       string `json:"outputText"`
	CompletionReason string `json:"completionReason"`
}

func (wrapper InvokeModelWrapper) InvokeTitanText(prompt string) (string, error) {
	modelId := "amazon.titan-text-express-v1"

	body, err := json.Marshal(TitanTextRequest{
		InputText: prompt,
		TextGenerationConfig: TextGenerationConfig{
			Temperature:   0,
			TopP:          1,
			MaxTokenCount: 4096,
		},
	})

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

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

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

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

	return response.Results[0].OutputText, nil
}

将您的第一个提示发送到 Amazon Titan Text。

// Send a prompt to Amazon Titan Text and print the response.
public class TextQuickstart {

    public static void main(String[] args) {

        // Create a Bedrock Runtime client in the AWS Region of your choice.
        var client = BedrockRuntimeClient.builder()
                .region(Region.US_EAST_1)
                .build();

        // You can replace the modelId with any other Titan Text Model. All current model IDs
        // are documented at https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html
        var modelId = "amazon.titan-text-premier-v1:0";

        // Define the prompt to send.
        var prompt = "Describe the purpose of a 'hello world' program in one line.";

        // Create a JSON payload using the model's native structure.
        var nativeRequest = new JSONObject().put("inputText", prompt);

        // Encode and send the request.
        var response = client.invokeModel(req -> req
                .body(SdkBytes.fromUtf8String(nativeRequest.toString()))
                .modelId(modelId));

        // Decode the response body.
        var responseBody = new JSONObject(response.body().asUtf8String());

        // Extract and print the response text.
        var responseText = responseBody.getJSONArray("results").getJSONObject(0).getString("outputText");

        System.out.println(responseText);
    }
}

使用系统提示符和其他推理参数调用 Titan Text。

    /**
     * Invoke Titan Text with a system prompt and additional inference parameters,
     * using Titan's native request/response structure.
     *
     * @param userPrompt   - The text prompt to send to the model.
     * @param systemPrompt - A system prompt to provide additional context and instructions.
     * @return The {@link JSONObject} representing the model's response.
     */
    public static JSONObject invokeWithSystemPrompt(String userPrompt, String systemPrompt) {

        // Create a Bedrock Runtime client in the AWS Region of your choice.
        var client = BedrockRuntimeClient.builder()
                .region(Region.US_EAST_1)
                .build();

        // Set the model ID, e.g., Titan Text Premier.
        var modelId = "amazon.titan-text-premier-v1:0";

        /* Assemble the input text.
         * For best results, use the following input text format:
         *     {{ system instruction }}
         *     User: {{ user input }}
         *     Bot:
         */
        var inputText = """
                %s
                User: %s
                Bot: 
                """.formatted(systemPrompt, userPrompt);

        // Format the request payload using the model's native structure.
        var nativeRequest = new JSONObject()
                .put("inputText", inputText)
                .put("textGenerationConfig", new JSONObject()
                        .put("maxTokenCount", 512)
                        .put("temperature", 0.7F)
                        .put("topP", 0.9F)
                )
                .toString();

        // Encode and send the request.
        var response = client.invokeModel(request -> {
            request.body(SdkBytes.fromUtf8String(nativeRequest));
            request.modelId(modelId);
        });

        // Decode the native response body.
        var nativeResponse = new JSONObject(response.body().asUtf8String());

        // Extract and print the response text.
        var responseText = nativeResponse.getJSONArray("results").getJSONObject(0).getString("outputText");
        System.out.println(responseText);

        // Return the model's native response.
        return nativeResponse;
    }

使用对话历史记录，使用 Titan Text 创建类似聊天的体验。

    /**
     * Create a chat-like experience with a conversation history, using Titan's native
     * request/response structure.
     *
     * @param prompt       - The text prompt to send to the model.
     * @param conversation - A String representing previous conversational turns in the format
     *                     User: {{ previous user prompt}}
     *                     Bot: {{ previous model response }}
     *                     ...
     * @return The {@link JSONObject} representing the model's response.
     */
    public static JSONObject invokeWithConversation(String prompt, String conversation) {

        // Create a Bedrock Runtime client in the AWS Region of your choice.
        var client = BedrockRuntimeClient.builder()
                .region(Region.US_EAST_1)
                .build();

        // Set the model ID, e.g., Titan Text Premier.
        var modelId = "amazon.titan-text-premier-v1:0";

        /* Append the new prompt to the conversation.
         * For best results, use the following text format:
         *     User: {{ previous user prompt}}
         *     Bot: {{ previous model response }}
         *     User: {{ new user prompt }}
         *     Bot: """
         */
        conversation = conversation + """
                %nUser: %s
                Bot: 
                """.formatted(prompt);

        // Format the request payload using the model's native structure.
        var nativeRequest = new JSONObject().put("inputText", conversation);

        // Encode and send the request.
        var response = client.invokeModel(request -> {
            request.body(SdkBytes.fromUtf8String(nativeRequest.toString()));
            request.modelId(modelId);
        });

        // Decode the native response body.
        var nativeResponse = new JSONObject(response.body().asUtf8String());

        // Extract and print the response text.
        var responseText = nativeResponse.getJSONArray("results").getJSONObject(0).getString("outputText");
        System.out.println(responseText);

        // Return the model's native response.
        return nativeResponse;
    }

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,
} from "@aws-sdk/client-bedrock-runtime";

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

/**
 * Invokes an Amazon Titan Text generation model.
 *
 * @param {string} prompt - The input text prompt for the model to complete.
 * @param {string} [modelId] - The ID of the model to use. Defaults to "amazon.titan-text-express-v1".
 */
export const invokeModel = async (
  prompt,
  modelId = "amazon.titan-text-express-v1",
) => {
  // Create a new Bedrock Runtime client instance.
  const client = new BedrockRuntimeClient({ region: "us-east-1" });

  // Prepare the payload for the model.
  const payload = {
    inputText: prompt,
    textGenerationConfig: {
      maxTokenCount: 4096,
      stopSequences: [],
      temperature: 0,
      topP: 1,
    },
  };

  // Invoke the model 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.
  const decodedResponseBody = new TextDecoder().decode(apiResponse.body);
  /** @type {ResponseBody} */
  const responseBody = JSON.parse(decodedResponseBody);
  return responseBody.results[0].outputText;
};

// Invoke the function if this file was run directly.
if (process.argv[1] === fileURLToPath(import.meta.url)) {
  const prompt =
    'Complete the following in one sentence: "Once upon a time..."';
  const modelId = FoundationModels.TITAN_TEXT_G1_EXPRESS.modelId;
  console.log(`Prompt: ${prompt}`);
  console.log(`Model ID: ${modelId}`);

  try {
    console.log("-".repeat(53));
    const response = await invokeModel(prompt, modelId);
    console.log(response);
  } catch (err) {
    console.log(err);
  }
}

使用调用模型 API 发送短信。

# Use the native inference API to send a text message to Amazon Titan Text.

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., Titan Text Premier.
model_id = "amazon.titan-text-premier-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 = {
    "inputText": prompt,
    "textGenerationConfig": {
        "maxTokenCount": 512,
        "temperature": 0.5,
    },
}

# 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["results"][0]["outputText"]
print(response_text)