Add Intel's GPU / XPU support to TIMM's validation script

[cfgfung]

Background

This PR is mainly about increasing the diversity of accelerators. Intel has a new GPU released last year and it is promoting to different platforms. This is a first step to incorporate the new hardware into TIMM. More optimizations and engineering efforts will be applied.

Related materials:
https://pytorch.org/tutorials/recipes/intel_extension_for_pytorch.html
https://github.com/intel/intel-extension-for-pytorch/tree/xpu-master

Environment

PyTorch version: 2.1.0a0+cxx11.abi
PyTorch CXX11 ABI: Yes
IPEX version: 2.1.10+xpu
OS: Ubuntu 22.04.3 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0

GPU: Intel(R) Data Center GPU Max 1550
Dataset: Imagenet-1K validation

Result

Without autocast

python validate.py --data-dir ../datasets/ --dataset hfds/imagenet-1k --split validation --device xpu --model resnet50.a1_in1k --batch-size 256 --pretrained

With autocast:

python validate.py --data-dir ../datasets/ --dataset hfds/imagenet-1k --split validation --device xpu --model resnet50.a1_in1k --batch-size 256 --pretrained --amp

[rwightman]

@cfgfung Hi, thanks for the PR, I have to come up with a better plan for handling the growing number of accel differences, with #2109 etc need a better way to centralize this code and support easy addition without needing checks and imports in every script...

I had an idea in the works in the bitsandtpu branch for TPU support, but haven't had a chance to unify that here with what I've learned since.

[cfgfung]

Hi Ross,

Thanks for the prompt review! Do you have any design/draft plan in your mind? We can have some discussions on unifying the interface/APIs. I am working on Intel's GPU and Gaudi. Might be able to provide some insights for those. Please feel free to contact me through email, kwun.fung.lau@intel.com

Related information:
https://www.intc.com/news-events/press-releases/detail/1689/intel-unleashes-enterprise-ai-with-gaudi-3-ai-open-systems
https://www.intel.com/content/www/us/en/products/details/discrete-gpus/data-center-gpu/max-series.html