DiffBIR icon indicating copy to clipboard operation
DiffBIR copied to clipboard

Training duration on A100

Open kirrukirru opened this issue 1 year ago • 3 comments

I started training with about 2000 images with a batch size of 10.

  1. How long does traning take for 2000 images set? Currently, it is at about 800 Epoch and each Epoch taking about 2.5 mins. Doesn't show total Epochs to process.
  2. I can see files like step=49999.ckpt etc. are created after every 10000 steps. If the training is stopped and started again, will it resume from where it was stopped?
  3. Can the training be done on CPU only?

Thanks, Kiran

kirrukirru avatar Sep 19 '23 10:09 kirrukirru