DAWNBench Imagenet--> Training face recognition model

First off, congrats to the DAWNBench team (@jeremy, @sgugger,@asparagui, & Andrew Shaw) on their impressive results!!

I’d like to replicate the approach to pre-train a face recognition model from scratch. I was wondering whether the code(s) are available for use. Could someone please confirm whether this is the correct repo?

Also, it would be great to know whether anyone has attempted to train Imagenet on a single GPU. (I have a GTX 1080).

  1. thanks! :smiley:
  2. that is the correct repo/scripts.
  3. we used volta gpu’s + half precision mode, which may not translate to your gpu. but in theory, the above scripts should work on a single machine. been thinking about trying it myself as well! :nerd_face:
  4. if you have more questions, send away! happy to help!