Is there a way to train in a way that consumes less VRAM?
yutaokuyama opened this issue · comments
Hi, thanks for your excellent research.
I would like to do a new training to compare your method with the method I am researching.
Due to limited computing resources, I would like to keep the VRAM consumption to about 48 GB, is it possible to do so through a configuration item?
Thank you in advance.
You can try using smaller N_samples
and patch.size
in the config file for less VRAM consumption, for instance N_samples: 64
size: 256
. Note the performance is not guaranteed for this configuration.
Thank you for your response. I really appreciate it.