J
jmustard
Guest
Hello,
For a while I have only been using DFL's H64 Model. I recently upgraded by GPU to a RTX 2070 with 8 GBs of VRAM. I am trying to learn how to use the SAE model but more specifically, create a model that will get the most out of my GPU. I have experimented with a resolution of 160 and a batch size of 4 (everything else default). This seems to be the max resolution I can get before going OOM. Should I be focusing on using the max resolution I can get? Or should I focus more on the batch sizes, dims, optimizer mode, multiscale decoder, etc...
For a while I have only been using DFL's H64 Model. I recently upgraded by GPU to a RTX 2070 with 8 GBs of VRAM. I am trying to learn how to use the SAE model but more specifically, create a model that will get the most out of my GPU. I have experimented with a resolution of 160 and a batch size of 4 (everything else default). This seems to be the max resolution I can get before going OOM. Should I be focusing on using the max resolution I can get? Or should I focus more on the batch sizes, dims, optimizer mode, multiscale decoder, etc...