AI-900 Practice Question: Describe features of generative AI workloads on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  The maximum number of paragraphs in the generated response

  Why it's wrong here

  The number of paragraphs in generated text is a formatting or length property, influenced by parameters such as max_tokens, stop sequences, or the prompt's instruction, not by top_p. top_p operates at the token level, modifying the probability distribution from which each next token is drawn, and has no direct control over document structure, paragraph breaks, or response length. Confusing these conflates content length with the statistical sampling mechanism that shapes word choice.

• ✓

  A sampling method that restricts token selection to the most probable token set

  Why this is correct

  top_p, also called nucleus sampling, is a sampling method where at every decoding step the model sorts all candidate tokens by descending probability and selects the smallest set whose cumulative probability reaches p (e.g., 0.95). It then samples only from that high-probability token set, proportionally to their probabilities. This restricts token selection to the most probable subset while still allowing some stochastic variation, controlling output diversity more directly than a fixed top-k cutoff.

• ✗

  A parameter that sets the minimum response quality threshold

  Why it's wrong here

  There is no generation parameter that sets an explicit 'minimum response quality threshold'; quality is measured afterward via human judgment, automated metrics, or downstream task performance. top_p instead controls the statistical diversity of sampling by choosing tokens from a dynamic nucleus of the probability distribution. It neither guarantees factual correctness nor enforces a baseline for linguistic quality, so this description mischaracterizes both its purpose and its mechanism.

• ✗

  The priority level of the API request in a queue

  Why it's wrong here

  Request priority in a queue is an API-management concern, typically handled by gateways, rate limiters, or message-queue ordering to decide which incoming call gets processed first. It has nothing to do with how the language model selects output tokens. top_p (nucleus sampling) is a decoding parameter that shapes response diversity by controlling which subset of vocabulary tokens can be sampled at each generation step, not the queuing or scheduling of API requests.

Go deeper

Related to this question