Junyu Chen

Hi Stephen, You can refer to the readme here to extract latent data. https://github.com/mit-han-lab/efficientvit/blob/5dd097d341a9cb2649733285d57e1efe6f35c0bd/applications/dc_ae/README.md?plain=1#L190

Hi Stephen, We tried some images from the RUGD dataset and observed that our autoencoders worked well. Here are some examples. The left part is the original image and the...

@StephenYangjz Thanks for your suggestion. The training command for DiT-XL on ImageNet 512x512 is added [here](https://github.com/mit-han-lab/efficientvit/blob/5b8f25e84bdf388128266610c117b263d5237e8d/applications/dc_ae/README.md?plain=1#L205) and [here](https://github.com/mit-han-lab/efficientvit/blob/5b8f25e84bdf388128266610c117b263d5237e8d/applications/dc_ae/README.md?plain=1#L227). If you want to finetune from the imagenet pretrained checkpoint, you can...

Hi @ZouYa99 , I ran the same code (except replacing `/data/dc-ae-f32c32-sana-1.0` with `mit-han-lab/dc-ae-f32c32-sana-1.0`), and the result is normal. I'm not sure if there is something wrong with this local checkpoint....

@ZouYa99 I'm still not sure the reason for this NAN issue. Maybe something is wrong with the environment. Since our models are merged into diffusers, I would recommend you give...

OK. I can also help to verify the correctness.

Hi @a-r-r-o-w, I have a conversion script, but it is based on the original codebase. I think giving you converted checkpoints may be better. Where do you think should I...

Sure. Here is the conversion script, which should be able to run in the [efficientvit](https://github.com/mit-han-lab/efficientvit) repo. ``` python import os import sys from dataclasses import dataclass from omegaconf import OmegaConf,...

My current conversion script is supposed to support all the checkpoints. Also, these checkpoints don't differ a lot. I think it won't lead to much more additional effort to support...

@a-r-r-o-w Hi Aryan, may I have a question here? I noticed that only `block_out_channels` are shared between the encoder and the decoder, while all other args like `block_types`, `layers_per_block`, and...