AWS Certified AI Practitioner (AIF-C01) free question 4 / 10 · Domain 2: Fundamentals of GenAI
A team wants to generate summaries from tens of thousands of product descriptions overnight, and the responses do not need to come back right away. Which method matches the Amazon Bedrock batch inference page?
- Run a batch inference job on a model with Provisioned Throughput purchased to secure overnight capacity
- Put the input files in S3, have a batch inference job process them asynchronously, and receive the results in S3
- Use tool calls inside the batch inference job so each summary queries the product database
- Send the inputs together to the CountTokens API, get the summaries and token counts at once, and save them to S3
Show the answer and explanation
Answer: B
Why this is correct
The page says batch inference lets you submit multiple prompts and generate responses asynchronously. You put a file that defines the input in an S3 bucket, send a batch inference request specifying the bucket, and when the job finishes you receive the output files from S3. It can be used to improve inference performance on large datasets, which suits large volumes that do not need an immediate response.
Why the other options are wrong
"Send the inputs together to the CountTokens API, get the summaries and token counts at once, and save them to S3" … CountTokens is an API that returns the number of input tokens before you send inference (see the page on counting tokens). It does not return inference results such as summaries.
"Run a batch inference job on a model with Provisioned Throughput purchased to secure overnight capacity" … A Note on the page says batch inference is not supported for provisioned models.
"Use tool calls inside the batch inference job so each summary queries the product database" … A Note on the page says batch inference does not support tool use (function calling) or structured outputs. Each input record is processed independently, with no round trips to the model.
Key point
Exam Guide objective 2.3.4 lists responsiveness as one of the cost tradeoffs. When you do not need an immediate response, processing in bulk and asynchronously is easier to handle than waiting on interactive calls one at a time. The page points you to the Amazon Bedrock pricing page for batch inference pricing, and job completion can be notified through EventBridge.