ai-forever / MoVQGAN

MoVQGAN - model for the image encoding and reconstruction
148 stars 11 forks source link

cuda out of memory #7

Open Droliven opened 8 months ago

Droliven commented 8 months ago

I modified the 67M version configs to train the network with 384*384 imgs, the size of z changed to 48*48*4. However it shown "cuda out of memory", even on an A100*80G GPU