Nina

Face the same problem, did you solve it?

> I've also tried multi-gpus and had same issue with you. > Then I found the biggest overhead is in GE2ELoss part. Especially get cosine similarity matrix and calculate loss...

> Your loss is increasing instead of decreasing... Have you used the trick to add a tiny value to the embeddings containing lots of zero values? And try to reduce...

My machine has enough free memory, but parse() cost long time about 200ms. Do you find how to speed up the parse() interface? @cmathx

The default target_loss is loss_0, loss_0 is discriminator loss, but default steps_to_start_discriminator is 200000 in UnivnetConfig and MultibandMelganConfig, so target loss not found error appear. And set target_loss='loss_1' in the...