Cannot train pytorch version on multi-gpu

thuml / CDAN

Code release for "Conditional Adversarial Domain Adaptation" (NIPS 2018)

398 stars 92 forks source link

Cannot train pytorch version on multi-gpu #23

Open chester256 opened 4 years ago

chester256 commented 4 years ago

When trained on multi-gpu, the transfer loss (CDAN + Entropy loss) is very low and the test accuracy alway decreases. Dataset: visda2017 tested on Ubuntu 18.04, Pytorch version: 1.2.0 / 1.4.0, GPU: GTX 1080 / RTX 2080Ti

shuowoshishui commented 4 years ago

我也会出现这样的情况