When using the openvino backend, do not look for an nvidia gpu.

[chrismuzyn]

Description

This fixes the situation where the openvino backend is asked for using the --use-openvino flag, but an nvidia gpu is still detected and used. Issue reference: #3920

Notes

This fix does not break Intel ARC A series gpus using the Opvenvino backend. This is in spite of the device variable being set to cpu.

Environment and Testing

I have tested this on two Fedora 41 systems running Stable Diffusion 3.5 Medium due to memory constraints on the latter system.

XEON 8468 + Nvidia RTX 4000 Ada: The Openvino backend properly uses cpu only, whereas before it tried to swap to the Nvidia backend and use the Nvidia gpu.

Ryzen 5600X + Intel Arc A380: The Openvino backend picked up the fact that the A380 was available as an Openvino device (note the device variable quirk above). Log messages show:

Torch parameters: backend=openvino device=GPU config=Auto dtype=torch.float32 context=no_grad nohalf=True nohalfvae=True upcast=False deterministic=False tunable=[False, False] fp16=fail bf16=fail optimization="Scaled-Dot-Product"

[vladmandic]

@chrismuzyn thnx - just for future reference - any pr should always target dev branch, never master directly.