Question 310
A provisioned model deployment is sized from average tokens per second, but the application experiences a predictable morning burst of large requests. What load test should guide sizing?
Explanation
Correct Answer
Provisioned throughput must handle the token mix and concurrent peak, not just daily averages.
Incorrect Answers
Idle health cannot prove peak behavior.
Training loss is not a serving quota metric.
A one-token assumption ignores workload variability.