VQ-Diffusion icon indicating copy to clipboard operation
VQ-Diffusion copied to clipboard

any training scripts for the pretraining the CC datasets?

Open runzeer opened this issue 2 years ago • 2 comments

runzeer avatar Aug 12 '22 03:08 runzeer

Also, I want to know the training cost when training on the larger dataset like LAION. Could you tell me the training epochs when training the larger models and the relevant costage about the GPU cards and the whole training time? Thanks a lot!

runzeer avatar Aug 12 '22 08:08 runzeer

In VQ-Diffusion, the largest dataset is LAION-person, which contains about 42M images, we trained it for 2 epochs. In improved VQ-Diffusion, the largest dataset is ITHQ dataset, which contains about 200M images, we also trained it for about 2 epochs on our large model, it cost about 96 V100s for a month.

cientgu avatar Sep 02 '22 09:09 cientgu