Yuqian Hong

What is "collapse"?

I got it. I appreciate yur work.

channel size maybe?

> Yeah, as you said, I reset the LLaMA-2 7B Path in another machine. It's work  Hello, I meet some problem about this code. Did you run the code...

https://github.com/lavinal712/AutoencoderKL/tree/dc-ae

This code supports training from scratch or fine-tuning the model, and it supports loading existing weight files without requiring conversion. https://github.com/lavinal712/AutoencoderKL/tree/dc-ae

As far as I know, AutoencoderKL is more of an AE (Autoencoder) than a VAE (Variational Autoencoder). Therefore, in the original code, the difference between L1 and L2 is not...

The difference between AutoencoderKL and a typical VAE is that you cannot actually sample from a Gaussian distribution and input it into the decoder to generate an image. AutoencoderKL relies...

Transfusion is inherently difficult to train; what you're essentially training is actually a DiT model.

I think it use diffusion method (or flow matching) in sampling. Am I wrong?