179 questions
No questions match those filters.
← All questions
System DesignMedium· Parallelism· Infrastructure· Inference
Why does aspect ratio matter beyond model quality?
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansWidth (dmodel) determines the feasibility of tensor parallelism because weight matrices are partitioned along the hidden dimension. Depth (nlayers) determines the number of pipeline stages and the granularity of those stages. These architectural choices interact directly with the networking topology of the training cluster, often becoming the primary bottleneck for cost and performance before model quality differences are even observed.