search

yccyenchicheng / SDFusion

MIT License
379 stars 32 forks source link

Computational resource. #6

Closed songwei100110 closed 1 year ago

songwei100110 commented 1 year ago

What your work need which GPU and how much time for training?

yccyenchicheng commented 1 year ago

we use V100 or A6000 to train our model. The memory usage are 32 GB or 48 GB per GPU. Training time varies by stages. VQVAE on all catetogries take around 7-10 days on one GPU. Diffusion model takes longer (also 7-14 days) and I don't quite remember, sorry!

We provide tensorboard to visualize the training progress.

RyanHTR commented 1 year ago

we use V100 or A6000 to train our model. The memory usage are 32 GB or 48 GB per GPU. Training time varies by stages. VQVAE on all catetogries take around 7~10 days on one GPU. Diffusion model takes longer (also 7~14 days) and I don't quite remember, sorry!

We provide tensorboard to visualize the training progress.

Do you mean it takes 7~14 days to train the model with one V100 GPU? Thanks.

yccyenchicheng commented 1 year ago

I update my comment. tilde becomes strike through in my original comment. Sorry for the confusion!