Add OpenVINO backend for torch.compile node

[openvino-dev-samples]

trafficstars

Features description

• To support both .safetensor model and LoRa weights with OpenVINO runtime.
• To support ALL of Intel CPU/GPU/NPU.

Installation

To enable this integration, you only need to install openvino runtime in advance:

pip install openvino
python3 main.py --cpu --use-pytorch-cross-attention

Test cases

• [x] model without lora node -> lora node enabled -> lora node disabled # current issue: /

• [x] model with lora node -> lora node disabled -> lora node enabled # current issue: it will need 2 warm-up model compilation (generation is slow) before fast generation with LoRA node

Screenshots

TorchCompileModel output has to be connected with the input of KSampler.

The following model has to be selected for checkpoint

https://github.com/comfyanonymous/ComfyUI/issues/2473