Training Code Release?
#6
by Changlinn - opened
Hi,
Thanks for the nice work! Do you have plan to release the training code or at least the GAN part?
I tried to copy the pipeline to train with WAN2.2 VAE, but the GAN loss suddenly converged in several iters. I thought the reason is the original VAE was too powerful to treat the discriminator. Thus, the VAE does not gain a lot through the GAN.