Textual inversion training notebook, 'AutoencoderKLOutput' object has no attribute 'sample'

[crypticsymmetry]

Describe the bug

Been getting this error was working last night... I've tried restarting/refreshing/New notebook, etc no luck. How do I fix this? thanks!

Reproduction

No response

Logs

AttributeError                            Traceback (most recent call last)
<ipython-input-21-7e0131ac8691> in <module>
      1 import accelerate
----> 2 accelerate.notebook_launcher(training_function, args=(text_encoder, vae, unet))

1 frames
<ipython-input-20-325883ee7544> in training_function(text_encoder, vae, unet)
     61             with accelerator.accumulate(text_encoder):
     62                 # Convert images to latent space
---> 63                 latents = vae.encode(batch["pixel_values"]).sample().detach()
     64                 latents = latents * 0.18215
     65 

AttributeError: 'AutoencoderKLOutput' object has no attribute 'sample'

System Info

colab notebook

[patrickvonplaten]

Good point, thanks for opening the issue :-)

We need to update this notebook for diffusers==0.3.0!

For now could you replace:

latents = vae.encode(batch["pixel_values"]).sample().detach()

with

latents = vae.encode(batch["pixel_values"]).latent_dist.sample().detach()

[crypticsymmetry]

ah gotcha. That works. thank you

[patrickvonplaten]

Official notebooks are now updated as well :-)