[アップデート] Amazon Nova Pro モデルがレイテンシー最適化推論をサポートしました

import json
import boto3

def extract_answer(response):
    """レスポンスから回答テキストを抽出する"""
    try:
        return response["output"]["message"]["content"][0]["text"]
    except (KeyError, IndexError) as e:
        return f"回答の抽出に失敗しました: {str(e)}"

def main():
    aws_region = "us-east-1"
    model_id = "us.amazon.nova-pro-v1:0"
    performance_config_latencies = ['standard', 'optimized']
    user_prompt = "日本で一番長い川は何ですか？その理由を教えてください"
    num_iterations = 5

    bedrock = boto3.client(
        'bedrock-runtime',
        region_name=aws_region
    )

    for performance_config_latency in performance_config_latencies:
        for i in range(num_iterations):
            try:
                response = bedrock.converse(
                    modelId=model_id,
                    messages=[
                        {
                            'role': 'user',
                            'content': [
                                {'text': user_prompt}
                            ]
                        }
                    ],
                    performanceConfig={
                        'latency': performance_config_latency
                    }
                )
            except Exception as e:
                print(f"エラーが発生しました: {str(e)}")

if __name__ == "__main__":
    main()

filter modelId = "us.amazon.nova-pro-v1:0" and performanceConfig.latency = "OPTIMIZED"
| stats avg(output.outputBodyJson.metrics.latencyMs) as AverageLatency

filter modelId = "us.amazon.nova-pro-v1:0" and performanceConfig.latency = "OPTIMIZED"
| stats avg(output.outputBodyJson.metrics.latencyMs) as AverageLatency