UCSB-NLP-Chang/DiffusionDisentanglement

GPU Memory needed for training

Closed this issue · 2 comments

zelaki commented

Thank you for your interesting work.
I have a question regarding the GPU memory size you used for training.
I tried to run disentangle.py with an NVIDIA RTX-3090 GPU with 24GB of memory and got an out-of-memory error.

Thank you in advance,
Thodoris

Hi Thodoris,

Thank you very much for your interest to our work! Unfortunately, as described here, this project requires a GPU memory of 48GB. You might want to try with a larger GPU. Sorry for the inconvenience!

Sincerely,
Qiucheng

zelaki commented

Hello Qiucheng,
I was looking for that info in the paper and didn't look at your README.
Thank you very much!
Thodoris