Questions about Commands for DLRM Training

[janghobaek2125]

With the command below, is preprocessing executed on the CPU while training runs on the GPU?

python dlrm_s_pytorch.py --arch-sparse-feature-size=64 --arch-mlp-bot="13-512-256-64" --arch-mlp-top="512-512-256-1" --max-ind-range=10000000 --data-generation=dataset --data-set=terabyte --raw-data-file=./input/day --processed-data-file=./input/terabyte_processed.npz --loss-function=bce --round-targets=True --learning-rate=0.1 --mini-batch-size=2048 --print-freq=1024 --print-time --use-gpu --dist-backend=nccl

Additionally, I would like to measure the preprocessing (CPU) and training times relative to the total training time.

Is there a way to view each of these times after training with the command above, or should I measure them individually using import time?