"slow_conv2d_cpu" not implemented for 'Half' how to fix this?

[dilectiogames]

I've followed this tutorial from the kohya_ss main page:

https://www.youtube.com/watch?v=TpuDOsuKIBo

it start the training then I get this error

RuntimeError: "slow_conv2d_cpu" not implemented for 'Half'

bellow full output

steps: 0%| | 0/240 [00:00<?, ?it/s]epoch 1/1 Traceback (most recent call last): File "E:\AI\kohya_ss\train_db.py", line 468, in <module> train(args) File "E:\AI\kohya_ss\train_db.py", line 267, in train latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample() File "E:\AI\kohya_ss\venv\lib\site-packages\diffusers\models\vae.py", line 566, in encode h = self.encoder(x) File "E:\AI\kohya_ss\venv\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) File "E:\AI\kohya_ss\venv\lib\site-packages\diffusers\models\vae.py", line 130, in forward sample = self.conv_in(sample) File "E:\AI\kohya_ss\venv\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) File "E:\AI\kohya_ss\venv\lib\site-packages\torch\nn\modules\conv.py", line 463, in forward return self._conv_forward(input, self.weight, self.bias) File "E:\AI\kohya_ss\venv\lib\site-packages\torch\nn\modules\conv.py", line 459, in _conv_forward return F.conv2d(input, weight, bias, self.stride, RuntimeError: "slow_conv2d_cpu" not implemented for 'Half' steps: 0%| | 0/240 [00:00<?, ?it/s]

is torch 2 btw

EDIT: I've disabled the "Mixed Precision" in the Training Parameters tab and now it fails without giving a precise error, only this at the end:

returned non-zero exit status 3221226356.

[dilectiogames]

tried with torch 1. same issue

RuntimeError: "slow_conv2d_cpu" not implemented for 'Half'

[dilectiogames]

I fixed this by choosing 0 instead of 1 when asked which GPU card to use.

what was confusing is that the Intel GPU is listed as GPU 0 and the Nvidia GPU is listed as 1 in windows.

even worse is that when kohya_ss starts it identifies the correct GPU and you think you got it right.

Instead either if you chose 0 or 1 it still show the correct GPU at startup but will fail to recognize that the GPU exist if you have chosen 1 in the setup

To fix, run the setup again, chose same options as before, it should be a fast process as all things are already installed. then when asked which of the GPUs to use, insert 0.

then the training will work.