Tutorial Lengkap AWS Bedrock: Managed Generative AI di AWS

Amazon Bedrock adalah layanan terkelola penuh yang menyediakan akses ke foundation models (FMs) dari perusahaan AI terkemuka melalui API terpadu. Layanan ini memungkinkan membangun aplikasi generative AI tanpa mengelola infrastruktur.

Mengapa AWS Bedrock?

• Multiple FMs: Akses Claude, Llama, Titan, dan lainnya
• Fully managed: Tidak perlu mengelola infrastruktur
• Aman: Privasi data dan dukungan VPC
• Customizable: Fine-tune model dengan data Anda
• Terintegrasi: Integrasi native dengan layanan AWS

• Anthropic Claude (Claude 3, Claude 2)
• Meta Llama 2
• Amazon Titan
• AI21 Labs Jurassic
• Cohere Command
• Stability AI (gambar)

Prerequisites

pip install boto3

Konfigurasi AWS CLI
aws configure

Enable akses model Bedrock di AWS Console

Quick Start

1. Basic Text Generation

import boto3
import json

Buat Bedrock runtime client
bedrock = boto3.client(
    servicename="bedrock-runtime",

    regionname="us-east-1"
)

Panggil model Claude
def generatetext(prompt):

    body = json.dumps({
        "anthropicversion": "bedrock-2023-05-31",
        "maxtokens": 1024,

        "messages": [
            {"role": "user", "content": prompt}
        ]
    })

    response = bedrock.invokemodel(
        modelId="anthropic.claude-3-sonnet-20240229-v1:0",
        body=body
    )

    result = json.loads(response["body"].read())
    return result["content"][0]["text"]

Generate teks
response = generatetext("Jelaskan machine learning dengan bahasa sederhana.")

print(response)

def generatetextstreaming(prompt):
    body = json.dumps({
        "anthropicversion": "bedrock-2023-05-31",
        "maxtokens": 1024,

        "messages": [
            {"role": "user", "content": prompt}
        ]
    })

    response = bedrock.invokemodelwithresponsestream(

        modelId="anthropic.claude-3-sonnet-20240229-v1:0",
        body=body
    )

    for event in response["body"]:
        chunk = json.loads(event["chunk"]["bytes"])
        if chunk["type"] == "contentblockdelta":

            print(chunk["delta"]["text"], end="", flush=True)

generatetextstreaming("Tulis puisi pendek tentang AI.")

def invoketitan(prompt):
    body = json.dumps({
        "inputText": prompt,
        "textGenerationConfig": {
            "maxTokenCount": 1024,
            "temperature": 0.7,
            "topP": 0.9
        }
    })

    response = bedrock.invokemodel(

        modelId="amazon.titan-text-express-v1",
        body=body
    )

    result = json.loads(response["body"].read())
    return result["results"][0]["outputText"]

response = invoketitan("Apa itu cloud computing?")
print(response)

2. Meta Llama 2

def invokellama(prompt):
    body = json.dumps({
        "prompt": f"[INST] {prompt} [/INST]",

        "maxgenlen": 512,

        "temperature": 0.7,
        "topp": 0.9
    })

    response = bedrock.invokemodel(

        modelId="meta.llama2-70b-chat-v1",
        body=body
    )

    result = json.loads(response["body"].read())
    return result["generation"]

response = invokellama("Jelaskan neural networks.")
print(response)

def invokecohere(prompt):
    body = json.dumps({
        "prompt": prompt,
        "maxtokens": 500,
        "temperature": 0.7
    })

    response = bedrock.invokemodel(

        modelId="cohere.command-text-v14",
        body=body
    )

    result = json.loads(response["body"].read())