When applying deep learning to real-world data, how should Benchmarks be understood? Select the option consistent with standard deep learning theory.

A. Benchmarks replace training algorithms B. Benchmarks remove the need for test data C. Standard benchmarks help compare architectures on common tasks and datasets D. Benchmarks are private weights inside a model