Error occur when training toy example

[koojahyeok]

Hi, I'm trying to train examples using the configurations in configs/example_training.

When I use yaml files such as imagenet-f8_cond.yaml or txt2img-clipl.yaml, there are no issues.
However, when I try using a yaml file located in configs/example_training/toy, I encounter the following error: ' AttributeError: 'IdentityFirstStage' object has no attribute 'decoder'. Did you mean: 'decode'?'.

They occur in sgm/models/diffusion.py at line 126.
The current line is if isinstance(self.first_stage_model.decoder, VideoDecoder): which results in the error.

Examples such as imagenet or txt2img-clipl use sgm.models.autoencoder.AutoencoderKL, which has decoder attribute . However, the examples located in configs/example_training/toy use sgm.models.autoencoder.IdentityFirstStage, which lacks the decoder attribute.

I believe a possible solution is to modify the code at sgm/models/diffusion.py line 126 to the following:
if hasattr(self.first_stage_model, 'decoder') and isinstance(self.first_stage_model.decoder, VideoDecoder):.

Thank you!

[xiexh20]

check out this pull request: #245
it solves my problem.

[koojahyeok]

But it solve just for code running...
I think my idea would solve original problem.