[Help]: How to Textual Inversion Embeddings on CPU with a cuda enabled machine?

[Annuakin]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What happened?

Textual Inversion Error, Assst please

Steps to reproduce the problem

1. Go to .... Training
2. Press .... Train
3. ... Error

What should have happened?

Well it should begin.

Commit where the problem happens

Textual Inversion/Train

What platforms do you use to access the UI ?

Windows

What browsers do you use to access the UI ?

Google Chrome

Command Line Arguments

--use-cpu --precision full --no-half

List of extensions

N/A

Console logs

raceback (most recent call last):
  File "C:\Users\moham\stable-diffusion-webui-master\modules\textual_inversion\textual_inversion.py", line 503, in train_embedding
    scaler.scale(loss).backward()
  File "C:\Users\moham\stable-diffusion-webui-master\venv\lib\site-packages\torch\cuda\amp\grad_scaler.py", line 164, in scale
    assert outputs.is_cuda or outputs.device.type == 'xla'
AssertionError

Additional information

Help please, I used to be able to do this on my cpu on my other laptop..... but it didnt actually use my GPU i guess cause it was an AMD and it was 4 GB.....

[ClashSAN]

hi @Annuakin, I am able to train both textual inversion and LoRA's (unet only) in 4gb, using https://github.com/kohya-ss/sd-scripts

I suggest you use this repo instead, as it has 8bit training options. It should work with Ampere series gpus.

[Zueuk]

I am able to train both textual inversion and LoRA's (unet only) in 4gb, using https://github.com/kohya-ss/sd-scripts

does it actually work on 4 Gb VRAM, or you mean CPU? how long does it take?

[ClashSAN]

yes, ~25 min. it depends if you can use bitsandbytes and xformers well on your machine.

[catboxanon]

Closing as stale.