使用具有回應串流的調用模型API在 Amazon Bedrock 上調用 Meta Llama 2

使用叫用模型API傳送文字訊息，並即時處理回應串流。

// Use the native inference API to send a text message to Meta Llama 2
// and print the response stream.

using System;
using System.IO;
using System.Text.Json;
using System.Text.Json.Nodes;
using Amazon;
using Amazon.BedrockRuntime;
using Amazon.BedrockRuntime.Model;

// Create a Bedrock Runtime client in the AWS Region you want to use.
var client = new AmazonBedrockRuntimeClient(RegionEndpoint.USEast1);

// Set the model ID, e.g., Llama 2 Chat 13B.
var modelId = "meta.llama2-13b-chat-v1";

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

// Embed the prompt in Llama 2's instruction format.
var formattedPrompt = $"<s>[INST] {prompt} [/INST]";

//Format the request payload using the model's native structure.
var nativeRequest = JsonSerializer.Serialize(new
{
    prompt = formattedPrompt,
    max_gen_len = 512,
    temperature = 0.5
});

// Create a request with the model ID and the model's native request payload.
var request = new InvokeModelWithResponseStreamRequest()
{
    ModelId = modelId,
    Body = new MemoryStream(System.Text.Encoding.UTF8.GetBytes(nativeRequest)),
    ContentType = "application/json"
};

try
{
    // Send the request to the Bedrock Runtime and wait for the response.
    var streamingResponse = await client.InvokeModelWithResponseStreamAsync(request);

    // Extract and print the streamed response text in real-time.
    foreach (var item in streamingResponse.Body)
    {
        var chunk = JsonSerializer.Deserialize<JsonObject>((item as PayloadPart).Bytes);
        var text = chunk["generation"] ?? "";
        Console.Write(text);
    }
}
catch (AmazonBedrockRuntimeException e)
{
    Console.WriteLine($"ERROR: Can't invoke '{modelId}'. Reason: {e.Message}");
    throw;
}

使用叫用模型API傳送文字訊息，並即時處理回應串流。

// Use the native inference API to send a text message to Meta Llama 2
// and print the response stream.

import org.json.JSONObject;
import org.json.JSONPointer;
import software.amazon.awssdk.auth.credentials.DefaultCredentialsProvider;
import software.amazon.awssdk.core.SdkBytes;
import software.amazon.awssdk.regions.Region;
import software.amazon.awssdk.services.bedrockruntime.BedrockRuntimeAsyncClient;
import software.amazon.awssdk.services.bedrockruntime.model.InvokeModelWithResponseStreamRequest;
import software.amazon.awssdk.services.bedrockruntime.model.InvokeModelWithResponseStreamResponseHandler;

import java.util.concurrent.ExecutionException;

import static software.amazon.awssdk.services.bedrockruntime.model.InvokeModelWithResponseStreamResponseHandler.Visitor;

public class Llama2_InvokeModelWithResponseStream {

    public static String invokeModelWithResponseStream() throws ExecutionException, InterruptedException {

        // Create a Bedrock Runtime client in the AWS Region you want to use.
        // Replace the DefaultCredentialsProvider with your preferred credentials provider.
        var client = BedrockRuntimeAsyncClient.builder()
                .credentialsProvider(DefaultCredentialsProvider.create())
                .region(Region.US_EAST_1)
                .build();

        // Set the model ID, e.g., Llama 2 Chat 13B.
        var modelId = "meta.llama2-13b-chat-v1";

        // The InvokeModelWithResponseStream API uses the model's native payload.
        // Learn more about the available inference parameters and response fields at:
        // https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-meta.html
        var nativeRequestTemplate = "{ \"prompt\": \"{{instruction}}\" }";

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

        // Embed the prompt in Llama 2's instruction format.
        var instruction = "<s>[INST] {{prompt}} [/INST]\\n".replace("{{prompt}}", prompt);

        // Embed the instruction in the the native request payload.
        var nativeRequest = nativeRequestTemplate.replace("{{instruction}}", instruction);

        // Create a request with the model ID and the model's native request payload.
        var request = InvokeModelWithResponseStreamRequest.builder()
                .body(SdkBytes.fromUtf8String(nativeRequest))
                .modelId(modelId)
                .build();

        // Prepare a buffer to accumulate the generated response text.
        var completeResponseTextBuffer = new StringBuilder();

        // Prepare a handler to extract, accumulate, and print the response text in real-time.
        var responseStreamHandler = InvokeModelWithResponseStreamResponseHandler.builder()
                .subscriber(Visitor.builder().onChunk(chunk -> {
                    // Extract and print the text from the model's native response.
                    var response = new JSONObject(chunk.bytes().asUtf8String());
                    var text = new JSONPointer("/generation").queryFrom(response);
                    System.out.print(text);

                    // Append the text to the response text buffer.
                    completeResponseTextBuffer.append(text);
                }).build()).build();

        try {
            // Send the request and wait for the handler to process the response.
            client.invokeModelWithResponseStream(request, responseStreamHandler).get();

            // Return the complete response text.
            return completeResponseTextBuffer.toString();

        } catch (ExecutionException | InterruptedException e) {
            System.err.printf("Can't invoke '%s': %s", modelId, e.getCause().getMessage());
            throw new RuntimeException(e);
        }
    }

    public static void main(String[] args) throws ExecutionException, InterruptedException {
        invokeModelWithResponseStream();
    }
}

使用叫用模型API傳送文字訊息，並即時處理回應串流。

// Send a prompt to Meta Llama 2 and print the response stream in real-time.

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

// Create a Bedrock Runtime client in the AWS Region of your choice.
const client = new BedrockRuntimeClient({ region: "us-west-2" });

// Set the model ID, e.g., Llama 2 Chat 13B.
const modelId = "meta.llama2-13b-chat-v1";

// Define the user message to send.
const userMessage =
  "Describe the purpose of a 'hello world' program in one sentence.";

// Embed the message in Llama 2's prompt format.
const prompt = `<s>[INST] ${userMessage} [/INST]`;

// Format the request payload using the model's native structure.
const request = {
  prompt,
  // Optional inference parameters:
  max_gen_len: 512,
  temperature: 0.5,
  top_p: 0.9,
};

// Encode and send the request.
const responseStream = await client.send(
  new InvokeModelWithResponseStreamCommand({
    contentType: "application/json",
    body: JSON.stringify(request),
    modelId,
  }),
);

// Extract and print the response stream in real-time.
for await (const event of responseStream.body) {
  /** @type {{ generation: string }} */
  const chunk = JSON.parse(new TextDecoder().decode(event.chunk.bytes));
  if (chunk.generation) {
    process.stdout.write(chunk.generation);
  }
}

// Learn more about the Llama 3 prompt format at:
// https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3/#special-tokens-used-with-meta-llama-3

使用叫用模型API傳送文字訊息，並即時處理回應串流。

# Use the native inference API to send a text message to Meta Llama 2
# and print the response stream.

import boto3
import json

from botocore.exceptions import ClientError

# 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., Llama 2 Chat 13B.
model_id = "meta.llama2-13b-chat-v1"

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

# Embed the prompt in Llama 2's instruction format.
formatted_prompt = f"<s>[INST] {prompt} [/INST]"

# Format the request payload using the model's native structure.
native_request = {
    "prompt": formatted_prompt,
    "max_gen_len": 512,
    "temperature": 0.5,
}

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

try:
    # Invoke the model with the request.
    streaming_response = client.invoke_model_with_response_stream(
        modelId=model_id, body=request
    )

    # Extract and print the response text in real-time.
    for event in streaming_response["body"]:
        chunk = json.loads(event["chunk"]["bytes"])
        if "generation" in chunk:
            print(chunk["generation"], end="")

except (ClientError, Exception) as e:
    print(f"ERROR: Can't invoke '{model_id}'. Reason: {e}")
    exit(1)