Amazon Bedrock Runtime examples using SDK for Go V2

package main

import (
	"context"
	"encoding/json"
	"flag"
	"fmt"
	"log"
	"os"
	"strings"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/config"
	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
)

// Each model provider defines their own individual request and response formats.
// For the format, ranges, and default values for the different models, refer to:
// https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters.html

type ClaudeRequest struct {
	Prompt            string `json:"prompt"`
	MaxTokensToSample int    `json:"max_tokens_to_sample"`
	// Omitting optional request parameters
}

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

// main uses the AWS SDK for Go (v2) to create an Amazon Bedrock Runtime client
// and invokes Anthropic Claude 2 inside your account and the chosen region.
// This example uses the default settings specified in your shared credentials
// and config files.
func main() {

	region := flag.String("region", "us-east-1", "The AWS region")
	flag.Parse()

	fmt.Printf("Using AWS region: %s\n", *region)

	ctx := context.Background()
	sdkConfig, err := config.LoadDefaultConfig(ctx, config.WithRegion(*region))
	if err != nil {
		fmt.Println("Couldn't load default configuration. Have you set up your AWS account?")
		fmt.Println(err)
		return
	}

	client := bedrockruntime.NewFromConfig(sdkConfig)

	modelId := "anthropic.claude-v2"

	prompt := "Hello, how are you today?"

	// Anthropic Claude requires you to enclose the prompt as follows:
	prefix := "Human: "
	postfix := "\n\nAssistant:"
	wrappedPrompt := prefix + prompt + postfix

	request := ClaudeRequest{
		Prompt:            wrappedPrompt,
		MaxTokensToSample: 200,
	}

	body, err := json.Marshal(request)
	if err != nil {
		log.Panicln("Couldn't marshal the request: ", err)
	}

	result, err := client.InvokeModel(ctx, &bedrockruntime.InvokeModelInput{
		ModelId:     aws.String(modelId),
		ContentType: aws.String("application/json"),
		Body:        body,
	})

	if err != nil {
		errMsg := err.Error()
		if strings.Contains(errMsg, "no such host") {
			fmt.Printf("Error: The Bedrock service is not available in the selected region. Please double-check the service availability for your region at https://aws.amazon.com/about-aws/global-infrastructure/regional-product-services/.\n")
		} else if strings.Contains(errMsg, "Could not resolve the foundation model") {
			fmt.Printf("Error: Could not resolve the foundation model from model identifier: \"%v\". Please verify that the requested model exists and is accessible within the specified region.\n", modelId)
		} else {
			fmt.Printf("Error: Couldn't invoke Anthropic Claude. Here's why: %v\n", err)
		}
		os.Exit(1)
	}

	var response ClaudeResponse

	err = json.Unmarshal(result.Body, &response)

	if err != nil {
		log.Fatal("failed to unmarshal", err)
	}
	fmt.Println("Prompt:\n", prompt)
	fmt.Println("Response from Anthropic Claude:\n", response.Completion)
}

Invoke multiple foundation models on Amazon Bedrock.

import (
	"context"
	"encoding/base64"
	"fmt"
	"log"
	"math/rand"
	"os"
	"path/filepath"
	"strings"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
	"github.com/awsdocs/aws-doc-sdk-examples/gov2/bedrock-runtime/actions"
	"github.com/awsdocs/aws-doc-sdk-examples/gov2/demotools"
)

// InvokeModelsScenario demonstrates how to use the Amazon Bedrock Runtime client
// to invoke various foundation models for text and image generation
//
// 1. Generate text with Anthropic Claude 2
// 2. Generate text with AI21 Labs Jurassic-2
// 3. Generate text with Meta Llama 2 Chat
// 4. Generate text and asynchronously process the response stream with Anthropic Claude 2
// 5. Generate an image with the Amazon Titan image generation model
// 6. Generate text with Amazon Titan Text G1 Express model
type InvokeModelsScenario struct {
	sdkConfig             aws.Config
	invokeModelWrapper    actions.InvokeModelWrapper
	responseStreamWrapper actions.InvokeModelWithResponseStreamWrapper
	questioner            demotools.IQuestioner
}

// NewInvokeModelsScenario constructs an InvokeModelsScenario instance from a configuration.
// It uses the specified config to get a Bedrock Runtime client and create wrappers for the
// actions used in the scenario.
func NewInvokeModelsScenario(sdkConfig aws.Config, questioner demotools.IQuestioner) InvokeModelsScenario {
	client := bedrockruntime.NewFromConfig(sdkConfig)
	return InvokeModelsScenario{
		sdkConfig:             sdkConfig,
		invokeModelWrapper:    actions.InvokeModelWrapper{BedrockRuntimeClient: client},
		responseStreamWrapper: actions.InvokeModelWithResponseStreamWrapper{BedrockRuntimeClient: client},
		questioner:            questioner,
	}
}

// Runs the interactive scenario.
func (scenario InvokeModelsScenario) Run(ctx context.Context) {
	defer func() {
		if r := recover(); r != nil {
			log.Printf("Something went wrong with the demo: %v\n", r)
		}
	}()

	log.Println(strings.Repeat("=", 77))
	log.Println("Welcome to the Amazon Bedrock Runtime model invocation demo.")
	log.Println(strings.Repeat("=", 77))

	log.Printf("First, let's invoke a few large-language models using the synchronous client:\n\n")

	text2textPrompt := "In one paragraph, who are you?"

	log.Println(strings.Repeat("-", 77))
	log.Printf("Invoking Claude with prompt: %v\n", text2textPrompt)
	scenario.InvokeClaude(ctx, text2textPrompt)

	log.Println(strings.Repeat("-", 77))
	log.Printf("Invoking Jurassic-2 with prompt: %v\n", text2textPrompt)
	scenario.InvokeJurassic2(ctx, text2textPrompt)

	log.Println(strings.Repeat("-", 77))
	log.Printf("Invoking Llama2 with prompt: %v\n", text2textPrompt)
	scenario.InvokeLlama2(ctx, text2textPrompt)

	log.Println(strings.Repeat("=", 77))
	log.Printf("Now, let's invoke Claude with the asynchronous client and process the response stream:\n\n")

	log.Println(strings.Repeat("-", 77))
	log.Printf("Invoking Claude with prompt: %v\n", text2textPrompt)
	scenario.InvokeWithResponseStream(ctx, text2textPrompt)

	log.Println(strings.Repeat("=", 77))
	log.Printf("Now, let's create an image with the Amazon Titan image generation model:\n\n")

	text2ImagePrompt := "stylized picture of a cute old steampunk robot"
	seed := rand.Int63n(2147483648)

	log.Println(strings.Repeat("-", 77))
	log.Printf("Invoking Amazon Titan with prompt: %v\n", text2ImagePrompt)
	scenario.InvokeTitanImage(ctx, text2ImagePrompt, seed)

	log.Println(strings.Repeat("-", 77))
	log.Printf("Invoking Titan Text Express with prompt: %v\n", text2textPrompt)
	scenario.InvokeTitanText(ctx, text2textPrompt)

	log.Println(strings.Repeat("=", 77))
	log.Println("Thanks for watching!")
	log.Println(strings.Repeat("=", 77))
}

func (scenario InvokeModelsScenario) InvokeClaude(ctx context.Context, prompt string) {
	completion, err := scenario.invokeModelWrapper.InvokeClaude(ctx, prompt)
	if err != nil {
		panic(err)
	}
	log.Printf("\nClaude     : %v\n", strings.TrimSpace(completion))
}

func (scenario InvokeModelsScenario) InvokeJurassic2(ctx context.Context, prompt string) {
	completion, err := scenario.invokeModelWrapper.InvokeJurassic2(ctx, prompt)
	if err != nil {
		panic(err)
	}
	log.Printf("\nJurassic-2 : %v\n", strings.TrimSpace(completion))
}

func (scenario InvokeModelsScenario) InvokeLlama2(ctx context.Context, prompt string) {
	completion, err := scenario.invokeModelWrapper.InvokeLlama2(ctx, prompt)
	if err != nil {
		panic(err)
	}
	log.Printf("\nLlama 2    : %v\n\n", strings.TrimSpace(completion))
}

func (scenario InvokeModelsScenario) InvokeWithResponseStream(ctx context.Context, prompt string) {
	log.Println("\nClaude with response stream:")
	_, err := scenario.responseStreamWrapper.InvokeModelWithResponseStream(ctx, prompt)
	if err != nil {
		panic(err)
	}
	log.Println()
}

func (scenario InvokeModelsScenario) InvokeTitanImage(ctx context.Context, prompt string, seed int64) {
	base64ImageData, err := scenario.invokeModelWrapper.InvokeTitanImage(ctx, prompt, seed)
	if err != nil {
		panic(err)
	}
	imagePath := saveImage(base64ImageData, "amazon.titan-image-generator-v1")
	fmt.Printf("The generated image has been saved to %s\n", imagePath)
}

func (scenario InvokeModelsScenario) InvokeTitanText(ctx context.Context, prompt string) {
	completion, err := scenario.invokeModelWrapper.InvokeTitanText(ctx, prompt)
	if err != nil {
		panic(err)
	}
	log.Printf("\nTitan Text Express    : %v\n\n", strings.TrimSpace(completion))
}

Use the Invoke Model API to send a text message.

import (
	"context"
	"encoding/json"
	"log"
	"strings"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
)

// InvokeModelWrapper encapsulates Amazon Bedrock actions used in the examples.
// It contains a Bedrock Runtime client that is used to invoke foundation models.
type InvokeModelWrapper struct {
	BedrockRuntimeClient *bedrockruntime.Client
}



// Each model provider has their own 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

type Jurassic2Request struct {
	Prompt      string  `json:"prompt"`
	MaxTokens   int     `json:"maxTokens,omitempty"`
	Temperature float64 `json:"temperature,omitempty"`
}

type Jurassic2Response struct {
	Completions []Completion `json:"completions"`
}
type Completion struct {
	Data Data `json:"data"`
}
type Data struct {
	Text string `json:"text"`
}

// Invokes AI21 Labs Jurassic-2 on Amazon Bedrock to run an inference using the input
// provided in the request body.
func (wrapper InvokeModelWrapper) InvokeJurassic2(ctx context.Context, prompt string) (string, error) {
	modelId := "ai21.j2-mid-v1"

	body, err := json.Marshal(Jurassic2Request{
		Prompt:      prompt,
		MaxTokens:   200,
		Temperature: 0.5,
	})

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

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

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

	var response Jurassic2Response
	if err := json.Unmarshal(output.Body, &response); err != nil {
		log.Fatal("failed to unmarshal", err)
	}

	return response.Completions[0].Data.Text, nil
}

Create an image with the Amazon Titan Image Generator.

import (
	"context"
	"encoding/json"
	"log"
	"strings"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
)

// InvokeModelWrapper encapsulates Amazon Bedrock actions used in the examples.
// It contains a Bedrock Runtime client that is used to invoke foundation models.
type InvokeModelWrapper struct {
	BedrockRuntimeClient *bedrockruntime.Client
}



type TitanImageRequest struct {
	TaskType              string                `json:"taskType"`
	TextToImageParams     TextToImageParams     `json:"textToImageParams"`
	ImageGenerationConfig ImageGenerationConfig `json:"imageGenerationConfig"`
}
type TextToImageParams struct {
	Text string `json:"text"`
}
type ImageGenerationConfig struct {
	NumberOfImages int     `json:"numberOfImages"`
	Quality        string  `json:"quality"`
	CfgScale       float64 `json:"cfgScale"`
	Height         int     `json:"height"`
	Width          int     `json:"width"`
	Seed           int64   `json:"seed"`
}

type TitanImageResponse struct {
	Images []string `json:"images"`
}

// Invokes the Titan Image model to create an image using the input provided
// in the request body.
func (wrapper InvokeModelWrapper) InvokeTitanImage(ctx context.Context, prompt string, seed int64) (string, error) {
	modelId := "amazon.titan-image-generator-v1"

	body, err := json.Marshal(TitanImageRequest{
		TaskType: "TEXT_IMAGE",
		TextToImageParams: TextToImageParams{
			Text: prompt,
		},
		ImageGenerationConfig: ImageGenerationConfig{
			NumberOfImages: 1,
			Quality:        "standard",
			CfgScale:       8.0,
			Height:         512,
			Width:          512,
			Seed:           seed,
		},
	})

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

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

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

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

	base64ImageData := response.Images[0]

	return base64ImageData, nil

}

Use the Invoke Model API to send a text message.

import (
	"context"
	"encoding/json"
	"log"
	"strings"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
)

// InvokeModelWrapper encapsulates Amazon Bedrock actions used in the examples.
// It contains a Bedrock Runtime client that is used to invoke foundation models.
type InvokeModelWrapper struct {
	BedrockRuntimeClient *bedrockruntime.Client
}



// 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(ctx context.Context, 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(ctx, &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
}

Invoke the Anthropic Claude 2 foundation model to generate text.

import (
	"context"
	"encoding/json"
	"log"
	"strings"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
)

// InvokeModelWrapper encapsulates Amazon Bedrock actions used in the examples.
// It contains a Bedrock Runtime client that is used to invoke foundation models.
type InvokeModelWrapper struct {
	BedrockRuntimeClient *bedrockruntime.Client
}



// 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(ctx context.Context, 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(ctx, &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
}

Use the Invoke Model API to send a text message and process the response stream in real-time.

import (
	"context"
	"encoding/json"
	"log"
	"strings"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
)

// InvokeModelWrapper encapsulates Amazon Bedrock actions used in the examples.
// It contains a Bedrock Runtime client that is used to invoke foundation models.
type InvokeModelWrapper struct {
	BedrockRuntimeClient *bedrockruntime.Client
}



// Each model provider defines their own individual request and response formats.
// For the format, ranges, and default values for the different models, refer to:
// https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters.html

type Request struct {
	Prompt            string  `json:"prompt"`
	MaxTokensToSample int     `json:"max_tokens_to_sample"`
	Temperature       float64 `json:"temperature,omitempty"`
}

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

// Invokes Anthropic Claude on Amazon Bedrock to run an inference and asynchronously
// process the response stream.

func (wrapper InvokeModelWithResponseStreamWrapper) InvokeModelWithResponseStream(ctx context.Context, prompt string) (string, error) {

	modelId := "anthropic.claude-v2"

	// Anthropic Claude requires you to enclose the prompt as follows:
	prefix := "Human: "
	postfix := "\n\nAssistant:"
	prompt = prefix + prompt + postfix

	request := ClaudeRequest{
		Prompt:            prompt,
		MaxTokensToSample: 200,
		Temperature:       0.5,
		StopSequences:     []string{"\n\nHuman:"},
	}

	body, err := json.Marshal(request)
	if err != nil {
		log.Panicln("Couldn't marshal the request: ", err)
	}

	output, err := wrapper.BedrockRuntimeClient.InvokeModelWithResponseStream(ctx, &bedrockruntime.InvokeModelWithResponseStreamInput{
		Body:        body,
		ModelId:     aws.String(modelId),
		ContentType: aws.String("application/json"),
	})

	if err != nil {
		errMsg := err.Error()
		if strings.Contains(errMsg, "no such host") {
			log.Printf("The Bedrock service is not available in the selected region. Please double-check the service availability for your region at https://aws.amazon.com/about-aws/global-infrastructure/regional-product-services/.\n")
		} else if strings.Contains(errMsg, "Could not resolve the foundation model") {
			log.Printf("Could not resolve the foundation model from model identifier: \"%v\". Please verify that the requested model exists and is accessible within the specified region.\n", modelId)
		} else {
			log.Printf("Couldn't invoke Anthropic Claude. Here's why: %v\n", err)
		}
	}

	resp, err := processStreamingOutput(ctx, output, func(ctx context.Context, part []byte) error {
		fmt.Print(string(part))
		return nil
	})

	if err != nil {
		log.Fatal("streaming output processing error: ", err)
	}

	return resp.Completion, nil

}

type StreamingOutputHandler func(ctx context.Context, part []byte) error

func processStreamingOutput(ctx context.Context, output *bedrockruntime.InvokeModelWithResponseStreamOutput, handler StreamingOutputHandler) (Response, error) {

	var combinedResult string
	resp := Response{}

	for event := range output.GetStream().Events() {
		switch v := event.(type) {
		case *types.ResponseStreamMemberChunk:

			//fmt.Println("payload", string(v.Value.Bytes))

			var resp Response
			err := json.NewDecoder(bytes.NewReader(v.Value.Bytes)).Decode(&resp)
			if err != nil {
				return resp, err
			}

			err = handler(ctx, []byte(resp.Completion))
			if err != nil {
				return resp, err
			}

			combinedResult += resp.Completion

		case *types.UnknownUnionMember:
			fmt.Println("unknown tag:", v.Tag)

		default:
			fmt.Println("union is nil or unknown type")
		}
	}

	resp.Completion = combinedResult

	return resp, nil
}

Use the Invoke Model API to send a text message.

import (
	"context"
	"encoding/json"
	"log"
	"strings"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
)

// InvokeModelWrapper encapsulates Amazon Bedrock actions used in the examples.
// It contains a Bedrock Runtime client that is used to invoke foundation models.
type InvokeModelWrapper struct {
	BedrockRuntimeClient *bedrockruntime.Client
}



// Each model provider has their own 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

type Llama2Request struct {
	Prompt       string  `json:"prompt"`
	MaxGenLength int     `json:"max_gen_len,omitempty"`
	Temperature  float64 `json:"temperature,omitempty"`
}

type Llama2Response struct {
	Generation string `json:"generation"`
}

// Invokes Meta Llama 2 Chat on Amazon Bedrock to run an inference using the input
// provided in the request body.
func (wrapper InvokeModelWrapper) InvokeLlama2(ctx context.Context, prompt string) (string, error) {
	modelId := "meta.llama2-13b-chat-v1"

	body, err := json.Marshal(Llama2Request{
		Prompt:       prompt,
		MaxGenLength: 512,
		Temperature:  0.5,
	})

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

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

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

	var response Llama2Response
	if err := json.Unmarshal(output.Body, &response); err != nil {
		log.Fatal("failed to unmarshal", err)
	}

	return response.Generation, nil
}