macos mps device support in benchmarks

🌟 Feature Description

macos mps device support in benchmarks

Motivation

It would be great to add support for macOS's MPS device.
By default, PyTorch models use CUDA (if available) or CPU, which isn’t very user-friendly for macOS devices with MPS acceleration.
The current device configuration code is:

self.device = "cuda:%s" % (GPU) if torch.cuda.is_available() and GPU >= 0 else "cpu"

To enable MPS support, we could modify the code as follows:

USE_CUDA = torch.cuda.is_available() and GPU >= 0
USE_MPS = torch.backends.mps.is_available()
self.device = torch.device(f'cuda:{GPU}' if USE_CUDA else ('mps' if USE_MPS else 'cpu'))

Alternatively, should we let users configure the device directly via a YAML parameter for more flexibility?

Start by locating the benchmark device configuration corresponding to the CUDA/CPU code shown in the issue, then inspect how benchmark devices are selected and whether YAML configuration already exists. Done means macOS benchmarks can select and use MPS without breaking CUDA or CPU fallback; the issue does not name a test file or benchmark entry point.