Benchmark the execution of a standard LLM model compiled through the torch-mlir pipeline without any new optimization passes against the current MPS backend of PyTorch.
Only eager mode is currently supported, MPS generation through torch.compile is not available (pytorch #150121)