Error running on cpu

[udiland]

Hi,

I try to run the program on ubuntu machine (ubntu on windows10) and its look like there is an issue with the torch_dtype argument.
After preprocessing I get this message:
'Pipelines loaded with torch_dtype=torch.float16 cannot run with cpu device. It is not recommended to move them to cpu as
running them will fail. Please make sure to use an accelerator to run the pipeline in inference, due to the lack of support forfloat16
operations on this device in PyTorch. Please, remove the torch_dtype=torch.float16 argument, or use another device for inference.'
And eventually the error is:
'RuntimeError: "LayerNormKernelImpl" not implemented for 'Half''

From what I found the dtype should be btype16 for cpu.
Thank for your help!

[iscas-zac]

Hello, I should have open an issue about this but simply forgot to.

I ran into the same problem a few days ago, and barely fixed it by changing the torch_dtype=torch.float16 to torch_dtype=torch.float32 in losses.py (I guess).

But of course it's very naive, and I didn't read through the paper. So I'm also looking forward to an official reply here. I think this deserves a patch update.