Textual inversion - Create Embedding Error

[oniris]

I found out my error was due to having the inpainting model loaded.

[laobishi080]

me too, the tip to me is this:
return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper__index_select)

what does it mean and what should i do .can anyone help me?

[CodeExplode]

It means that some of the information is in your system memory, and some is in your GPU memory, whereas the underlying pytorch system which handles machine learning stuff needs them to all be on one or the other. I'm not sure how it could have occurred though, and might have been fixed since then since I've made some embeddings in the latest version.

One thing I can think of, do you possibly have a low vector count (e.g. 1) and initialization text which maps to more than 1 vector? (e.g. if you put your initialization text in the txt2img prompt box, what number of vectors does it show in the small prompt count which shows up below?)

[cibernicola]

I have observed the same error, I have 6GB VRAM, if I start with any "optimization" parameter, the error described above appears, if I start without any parameter, I can start the training but the CUDA memory error appears.

edit: I just tried something, downgraded from 512x512 to 256x256 and I think I will be able to finish the training.

[Hagasus]

happens to me since today. Perhaps a bug...

[Decayo]

happened to me first, but after I
disable use artist.csv in Interrogate Options and without --medvram switch
it success create a new embedding pt file