A team uses End-to-End AutoML for tabular modeling and wants to shrink the final ensemble model to improve serving latency and lower inference cost. Which optional pipeline component addresses that need?

Question

Accepted Answer

C. condition-is-distill. In End-to-End AutoML, the optional distillation step creates a smaller version of the ensemble model, which reduces latency and cost for inference. The tuner searches architectures and hyperparameters, but it does not itself perform model size reduction after ensembling.

A team uses End-to-End AutoML for tabular modeling and wants to shrink the final ensemble model to improve serving latency and lower inference cost. Which optional pipeline component addresses that need?

Related Questions

Discussion