Test on VE settings

[chen-hao-chao]

Hi,
I really like your work and I wanted to use your method on VE settings.

I have tried this code on sub-VP settings, but have trouble under VE settings. Do you have any helpful comments on how I should use your code to sample from a pretrained VE model?

(Specifically, I would like to evaluate this code on the configuration "cifar10_deep_continuous_ve.py".)

Thanks!

[AlexiaJM]

Hi,

Any of these three lines should work:

score_sde_fast_sampling/experiments.sh

Line 129 in 4805a20

python main.py --config 'configs/ncsnpp/cifar10_deep_continuous_ve.py' --mode 'eval' --workdir 'cifar10_ncsnpp_deep_continuous' --config.sampling.predictor="adaptive" --config.sampling.corrector="none" --config.eval.num_samples=$total_lowres --config.eval.batch_size=$bs_lowres --config.sampling.abstol=0.0039 --config.sampling.reltol=1e-2

score_sde_fast_sampling/experiments.sh

Line 132 in 4805a20

python main.py --config 'configs/ncsnpp/cifar10_deep_continuous_ve.py' --mode 'eval' --workdir 'cifar10_ncsnpp_deep_continuous' --config.sampling.predictor="adaptive" --config.sampling.corrector="none" --config.eval.num_samples=$total_lowres --config.eval.batch_size=$bs_lowres --config.sampling.abstol=0.0039 --config.sampling.reltol=2e-2

score_sde_fast_sampling/experiments.sh

Line 135 in 4805a20

python main.py --config 'configs/ncsnpp/cifar10_deep_continuous_ve.py' --mode 'eval' --workdir 'cifar10_ncsnpp_deep_continuous' --config.sampling.predictor="adaptive" --config.sampling.corrector="none" --config.eval.num_samples=$total_lowres --config.eval.batch_size=$bs_lowres --config.sampling.abstol=0.0039 --config.sampling.reltol=5e-2

You just need to change --workdir to the directory of your pretrained models in ./checkpoints/.

Alexia

[chen-hao-chao]

Thank you so much.
I successfully sampled from a pretrained VE model using adaptive method under "cifar10_deep_continuous_ve.py" settings.