search

deepglugs / deep_imagen

scripts for running and training imagen-pytorch
38 stars 8 forks source link

Separate unets #4

Open krrishdholakia opened 2 years ago

krrishdholakia commented 2 years ago

is it possible to just train unet 1? When i try to set the num_unets to 0 (since that's what initializes unet 2), i get an error that image size length must == num unets

deepglugs commented 2 years ago

Use --num_unets=0 should work. I'll take a look

deepglugs commented 2 years ago

Should be fixed with https://github.com/deepglugs/deep_imagen/commit/e8b255f2fd55679cd1282901e40c7e3e261011cc

krrishdholakia commented 2 years ago

what does train_unet do? if i have 2 unets, do i need to set that to 2? or is it training the individual unet number specified?

deepglugs commented 2 years ago

with imagen-pytorch 1.x, you can only train 1 unet at a time. So train_unet specifies which to train. You'll want to train 1 for a while, then set to 2, train and so on.