Running inference on a trained model in AWS Clean Rooms ML

Members with the ability to run queries can also initiate inference job once the training job is complete. They pick the inference dataset against which they want to run inference and reference the trained model outputs that they’d like to run inference container with.

The member who will receive inference output must be granted the member ability CAN_RECEIVE_INFERENCE_OUTPUT.

Console

To create a model inference job (console)

1. Sign in to the AWS Management Console and open the AWS Clean Rooms console at https://console.aws.amazon.com/cleanrooms.

2. In the left navigation pane, choose Collaborations.

3. On the Collaborations page, choose the collaboration that contains the custom model that you want to create an inference job on.

4. After the collaboration opens, choose the ML Models tab, then choose your model from the Custom trained model table.

5. On the custom trained model details page, click Start inference job.

6. For Start inference job, for Inference job details, enter a Name and optional Description.

   Enter the following information:

   • Associated model algorithm - The associated model algorithm that is used during the inference job.

   • ML input channel details - The ML input channel that will provide the data for this inference job.

   • Transform resources - The compute instance that is used to perform the transform function of the inference job.

   • Output configuration - Who will receive the inference job output and the MIME type of the output.

   • Encryption - choose the Customize encryption settings to specify your own KMS key and related information. Otherwise, Clean Rooms ML will manage the encryption.

   • Transform job details - The maximum payload of the inference job, in MB.

   • Environment variables - Any environment variables necessary to access the inference job container image.

7. Choose Start inference job.

   The results are exported to the following path in the Amazon S3 location that was specified in the ML configuration: yourSpecifiedS3Path/collaborationIdentifier/trainedModelName/callerAccountId/jobName.

API

To create a model inference job (API)

Initiate the inference job by running the following code:

import boto3 
acr_ml_client= boto3.client('cleanroomsml')

acr_ml_client.start_trained_model_inference_job(
    name="inference_job",
    membershipIdentifier='membership_id',
    trainedModelArn='arn:aws:cleanrooms-ml:region:account:membership/membershipIdentifier/trained-model/identifier', 
    
    dataSource={
        "mlInputChannelArn": 'channel_arn_3'
    },
    resourceConfig={'instanceType': 'ml.m5.xlarge'},
    outputConfiguration={
        'accept': 'text/csv',
        'members': [
            {
                "accountId": 'member_account_id'
            }
        ]
    }
)

The results are exported to the following path in the Amazon S3 location that was specified in the ML configuration: yourSpecifiedS3Path/collaborationIdentifier/trainedModelName/callerAccountId/jobName.