MLA-C01 Practice Question 7: Asynchronous Inference Large Payloads
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 1 · question 7 of 10
A media company runs a video-analysis model whose input files are roughly 900 MB each and whose inference takes between 8 and 12 minutes per request. Requests arrive unpredictably, a few times per hour, and there are long idle stretches overnight. The calling application must submit a request to one endpoint API, receive an identifier immediately, and be notified when the result is ready. The company does not want to pay for inference compute while no requests are in flight.
Which SageMaker AI inference option should be used?
Show answer and explanation
Answer:
Asynchronous Inference is built for this exact profile: payloads up to 1 GB, processing times up to one hour, and near real-time rather than interactive latency. The caller places the payload in S3 and calls InvokeEndpointAsync with a pointer to it; SageMaker AI queues the request, returns an identifier and an output location immediately, writes the result back to S3, and can publish a success or error notification to SNS. Because the instance count can autoscale to zero, the company pays nothing during the idle overnight window.
Option A fails on every constraint. A real-time endpoint caps the request payload at 6 MB and the invocation at 60 seconds, so a 900 MB file and a 10-minute inference are both impossible, and a minimum capacity of one instance means paying around the clock.
Option B fails as well. Serverless Inference is a synchronous request/response option with a 6 GB memory ceiling and no GPU support; it cannot hold a connection open for the 8 to 12 minutes each request needs. Provisioned concurrency would also reintroduce the idle cost the company is trying to avoid.
Option D would eventually produce results, but it is the wrong shape. Each request would spin up fresh batch transform infrastructure with several minutes of startup overhead, the caller no longer talks to a single endpoint API, and the company takes on orchestration, retry, and notification plumbing that Asynchronous Inference provides natively.
Reference: Asynchronous inference