Exemplos de código para inferência em lote

Os exemplos de código neste capítulo mostram como criar um trabalho de inferência em lote, visualizar informações sobre ele e interrompê-lo. Selecione um idioma para ver um exemplo de código para ele:

Python

Crie um JSONL arquivo chamado abc.jsonl que contém pelo menos o número mínimo de registros (consulteCotas do Amazon Bedrock). Você pode usar o seguinte conteúdo como primeira linha e entrada:

{
    "recordId": "CALL0000001", 
    "modelInput": {
        "anthropic_version": "bedrock-2023-05-31", 
        "max_tokens": 1024,
        "messages": [ 
            { 
                "role": "user", 
                "content": [
                    {
                        "type": "text", 
                        "text": "Summarize the following call transcript: ..." 
                    } 
                ]
            }
        ]
    }
}

Crie um bucket S3 chamado amzn-s3-demo-bucket-input e faça o upload do arquivo para ele. Em seguida, crie um bucket S3 chamado amzn-s3-demo-bucket-output para gravar seus arquivos de saída. Execute o seguinte trecho de código para enviar um trabalho e obter o jobArn da resposta:

import boto3

bedrock = boto3.client(service_name="bedrock")

inputDataConfig=({
    "s3InputDataConfig": {
        "s3Uri": "s3://amzn-s3-demo-bucket-input/abc.jsonl"
    }
})

outputDataConfig=({
    "s3OutputDataConfig": {
        "s3Uri": "s3://amzn-s3-demo-bucket-output/"
    }
})

response=bedrock.create_model_invocation_job(
    roleArn="arn:aws:iam::123456789012:role/MyBatchInferenceRole",
    modelId="anthropic.claude-3-haiku-20240307-v1:0",
    jobName="my-batch-job",
    inputDataConfig=inputDataConfig,
    outputDataConfig=outputDataConfig
)

jobArn = response.get('jobArn')

Devolva o status do trabalho.

bedrock.get_model_invocation_job(jobIdentifier=jobArn)['status']

Listar trabalhos de inferência em lote que Failed.

bedrock.list_model_invocation_jobs(
    maxResults=10,
    statusEquals="Failed",
    sortOrder="Descending"
)

Interrompa o trabalho que você começou.

bedrock.stop_model_invocation_job(jobIdentifier=jobArn)

Java

package com.amazon.aws.sample.bedrock.inference;

import com.amazonaws.services.bedrock.AmazonBedrockAsync;
import com.amazonaws.services.bedrock.AmazonBedrockAsyncClientBuilder;
import com.amazonaws.services.bedrock.model.CreateModelInvocationJobRequest;
import com.amazonaws.services.bedrock.model.CreateModelInvocationJobResult;
import com.amazonaws.services.bedrock.model.GetModelInvocationJobRequest;
import com.amazonaws.services.bedrock.model.GetModelInvocationJobResult;
import com.amazonaws.services.bedrock.model.InvocationJobInputDataConfig;
import com.amazonaws.services.bedrock.model.InvocationJobOutputDataConfig;
import com.amazonaws.services.bedrock.model.InvocationJobS3InputDataConfig;
import com.amazonaws.services.bedrock.model.InvocationJobS3OutputDataConfig;
import com.amazonaws.services.bedrock.model.ListModelInvocationJobsRequest;
import com.amazonaws.services.bedrock.model.ListModelInvocationJobsResult;
import com.amazonaws.services.bedrock.model.StopModelInvocationJobRequest;
import com.amazonaws.services.bedrock.model.StopModelInvocationJobResult;

public class BedrockAsyncInference {
    private final AmazonBedrockAsync amazonBedrockAsyncClient = AmazonBedrockAsyncClientBuilder.defaultClient();
    public void createModelInvokeJobSampleCode() {
    
        final InvocationJobS3InputDataConfig invocationJobS3InputDataConfig = new InvocationJobS3InputDataConfig()
                .withS3Uri("s3://batch-input/abc.jsonl")
                .withS3InputFormat("JSONL");
    
        final InvocationJobInputDataConfig inputDataConfig = new InvocationJobInputDataConfig()
                .withS3InputDataConfig(invocationJobS3InputDataConfig);
    
        final InvocationJobS3OutputDataConfig invocationJobS3OutputDataConfig = new InvocationJobS3OutputDataConfig()
                .withS3Uri("s3://batch-output/");
    
        final InvocationJobOutputDataConfig invocationJobOutputDataConfig = new InvocationJobOutputDataConfig()
                .withS3OutputDataConfig(invocationJobS3OutputDataConfig);
    
    
        final CreateModelInvocationJobRequest createModelInvocationJobRequest = new CreateModelInvocationJobRequest()
                .withModelId("anthropic.claude-3-haiku-20240307-v1:0")
                .withJobName("unique-job-name")
                .withRoleArn("arn:aws:iam::123456789:role/bedrock-role")
                .withClientRequestToken("Client-token")
                .withInputDataConfig(inputDataConfig)
                .withOutputDataConfig(invocationJobOutputDataConfig);
    
    
        final CreateModelInvocationJobResult createModelInvocationJobResult = amazonBedrockAsyncClient
                .createModelInvocationJob(createModelInvocationJobRequest);
    
        System.out.println(createModelInvocationJobResult.getJobArn());
    }
    
    public void getModelInvokeJobSampleCode() {
        final GetModelInvocationJobRequest getModelInvocationJobRequest = new GetModelInvocationJobRequest()
                .withJobIdentifier("jobArn");
    
        final GetModelInvocationJobResult getModelInvocationJobResult = amazonBedrockAsyncClient
                .getModelInvocationJob(getModelInvocationJobRequest);
    
    }
    
    public void listModelInvokeJobSampleCode() {
        final ListModelInvocationJobsRequest listModelInvocationJobsRequest = new ListModelInvocationJobsRequest()
                .withMaxResults(10)
                .withNameContains("matchin-string");
    
        final ListModelInvocationJobsResult listModelInvocationJobsResult = amazonBedrockAsyncClient
                .listModelInvocationJobs(listModelInvocationJobsRequest);
    
    }
    
    public void stopModelInvokeJobSampleCode() {
        final StopModelInvocationJobRequest stopModelInvocationJobRequest = new StopModelInvocationJobRequest()
                .withJobIdentifier("jobArn");
    
        final StopModelInvocationJobResult stopModelInvocationJobResult = amazonBedrockAsyncClient
                .stopModelInvocationJob(stopModelInvocationJobRequest);
    
    }

}