Wang Qiang

> use torch.nn.DataParallel is enough

For previous diffusion models with 1D data, such as spatio-temporal sequences and 1D vector graphs, I think you need to at least modify the channel of Unet.

The number of training is related to your data set. You may need to observe loss, FID and generation quality. In my personal experience, too many training times will lose...

> @XDUWQ i brought back the old resnet blocks, so you can try turning on and off `ConvNext` using 'use_convnext` > > curious what you find is better! Thank you...

> Hi!, What loss (L1 and L2) should I expect for a properly trained model? And which loss usually performs better? In paper "https://arxiv.org/abs/2111.05826" has some discuss about L1 loss...

> https://github.com/znxlwm/pytorch-MNIST-CelebA-GAN-DCGAN/blob/c88438c7807326492757623bf9117ee2eb5de8bf/pytorch_MNIST_DCGAN.py#L210 可以

same question

`def slerp_theta(z1, z2, theta): return math.cos(theta) * z1 + math.sin(theta) * z2` 

I solve the same problem! ` model.load_state_dict( dist_util.load_state_dict(args.model_path, map_location="cpu"), strict=False )`

``` git clone https://github.com/LukasBommes/mv-extractor.git cd mv-extractor/ pip install . ```