b4nn3d

Only if you go at 1024*1024 resolution.. in the official repo there is a benchmark tool, you can try by yourself

I know that we can train unet alone, I'm asking if theorically we can get some advantages training only the text encoder ;)

> Sure thats no problem. I also think that we can update them one at a time, iteratively. This will get the best of both worlds: training both text, unet...

I launched the training with this. !python run_training.py --result-dir=results --data-dir=datasets --dataset=blow --config=config-f --total-kimg=12000 --mirror-augment=true --metric=none --min-h=3 --min-w=3 --res-log2=7

i got OOM when i was trying with a 512*512 dataset. this one was 384*384. in your example you train a 640x384 dataset, so i don't see how this could...

ok, it was a memory issue. trained for 220 ticks with your method