Hi there
I’m trying to train a VAE-style model using Pyro on an A100 GPU. I’ve got all the torch tensors on the gpu device, same with the nn.Modules that are components of the model and guide.
Training feels like a similar speed on the A100 compared to training on CPU of local device (M1 max, 64 gb ram).
Relative to Numpyro, I can find less document for Pyro and training on GPUs… can someone point to any docs or fatal errors that I might be making?
Thanks!