medium10 pts · first answer countsgpu-computingperformancepytorch
Logging slows the training loop
Adding print(loss.item()) and an accuracy check with .cpu() on every step makes training 25% slower. Why?
Discussion 0
Join free to post in the discussion.