AttributeError: 'UNet2DConditionModel' object has no attribute 'enable_gradient_checkpointing' about sd_dreambooth_extension

[entmike]

Update, it appears edcae96 broke this because of the diffusers change in requirements.txt. If I change it back to diffusers==0.7.2, this error goes away.

[rehaldinho64]

Changing the diffusers version didn't work for me. Switching from Python version 3.10 to 3.9 worked.

[d8ahazard]

I'm going to close this, as I believe this was related to issues with installing the plugin. Feel free to open a new issue with the Template if not so I can better diagnose the issue. Tx.

[Jonseed]

I'm also seeing this error on the latest commit 2eb125e, which installs diffusers 0.3.0? Or is that the server that is installing it? Going to try to change diffusers back to 0.7.2

[phutngo]

So what's the solution to this problem then?

Returning result: Exception training model: 'UNet2DConditionModel' object has no attribute 'params'

[travisbeard]

I am getting this as well. Using Python 3.10

[harrywang]

Changing the diffusers version didn't work for me. Switching from Python version 3.10 to 3.9 worked.

I ran into the same problem with Python 3.8.x and 3.9.11 worked.

[pmarreck]

I ran into a similar issue ("model stable-diffusion-xl-base-1.0 could not be loaded: 'UNet2DConditionModel' object has no attribute 'add_embedding'") and it got me here.

The fact that simply flipping between minor versions of Python fixes ANYTHING (and not even in a logical way! Like why would a new language version fix a bug that looks data- or code- related such as "invalid argument" or "no attribute" errors?) shows just how absolutely brittle Python is as a tech to build this stuff on. Just saying.

[threewood1]

I also got similar wrong like this:

AttributeError: 'UNetModel' object has no attribute 'enable_gradient_checkpointing'. Did you mean: '_return_value'?

and I changed python3.10.0 to python3.9.11, but it didn't work.