AWS Certified AI Practitioner (AIF-C01) · Objective 3.1
Choosing top-k
3.1Standard
Pass only the highest-ranked documents to generation.
Balance missed evidence and token cost.
Choose two evaluations.
Hint 1: Focus on the requirement
Tune the number of candidates against recall, evidence precision, duplicate context, token budget, and the quality after authorization filtering.
Hint 2: Key terms
Review top-k retrieval and recall.
Top-k
A setting that passes only the first k ranked search results to the next stage.
Recall
The proportion of relevant candidates that a retrieval process returns.
Evidence precision
The proportion of retrieved candidates that are actually relevant evidence.
Duplicate context
Repeated information that consumes capacity without adding distinct evidence.
× Incorrect
Compare recall, evidence accuracy, duplication, and token cost for different top-k values, including important documents after authorization filtering.
Detailed explanation
A
CorrectTest important documents that rank low and results after authorization filtering
Ranking and access effects are visible.
B
IncorrectA larger k always improves answers
Noise and cost also increase.
C
IncorrectJudge retrieval from one top result
Misses and diversity are not measured.
D
CorrectCompare recall, evidence accuracy, duplication, and tokens for several top-k values
Retrieval range and cost are measured.
Check official documentation
Compare the following points with the official documentation.
Path & Pass is an independent learning site and is not affiliated with any certification body. Questions are original practice content based on published objectives.