CPU only?

[Petomai]

Can this be used CPU only? I have only a cheap laptop, Intel UHD graphics and want to use CPU. When I try to start training with CPU selected for train device and temp device it says "huggingface/diffusers#254 .
Traceback (most recent call last):
File "C:\OneTrainer\modules\ui\TrainUI.py", line 522, in __training_thread_function
trainer.start()
File "C:\OneTrainer\modules\trainer\GenericTrainer.py", line 123, in start
self.model_setup.setup_model(self.model, self.config)
File "C:\OneTrainer\modules\modelSetup\StableDiffusionLoRASetup.py", line 147, in setup_model
model.optimizer = create.create_optimizer(
File "C:\OneTrainer\modules\util\create.py", line 338, in create_optimizer
optimizer = torch.optim.AdamW(
File "C:\OneTrainer\venv\lib\site-packages\torch\optim\adamw.py", line 68, in init
raise RuntimeError("fused=True requires all the params to be floating point Tensors of "
RuntimeError: fused=True requires all the params to be floating point Tensors of supported devices: ['cuda', 'xpu', 'privateuseone']."

I don't know what this means only that maybe somehow it is trying to find cuda but I don't have cuda device.

[mx]

I doubt you'll even be able to run the VAE with just your CPU. I do not believe what you are attempting is viable.

[Petomai]

I am not interested in creating VAE, only want to generate the Lora. There is no VAE data type selected in model so it shouldn't try to train VAE. I have chosen sd 1.5 Lora, device cpu but then there's the runtime error.

[mx]

The entire pipeline has to run as part of creating the lora, which includes the VAE. I doubt that your CPU on its own is even capable of running the VAE, the earliest part of the pipeline. You are welcome to try, you'll just have to do what the error message tells you to do -- turn off the fused option in the advanced optimizer settings.

[Petomai]

I got it to start training, it needed float32 instead of float16 for cpu training, and changed the optimizer to Lamb, turned attention to SDP but now it has some "not defined" exception in the code.
optim_func = str2optimizer32bit[optimizer_name][0]
"NameError: name 'str2optimizer32bit' is not defined"

Being poor kind of bites but it is what it is, thank you for the replies and help 😊

edit* I set the optimizer to Adam and did as you said to disable fused, it's now progressed further. it is very slow but slow is better than none 😁