Fix incorrect accelerator device handling for MPS in `TrainingArgumen…

…ts` (huggingface#31812)

* Fix wrong acclerator device setup when using MPS

* More robust TrainingArguments MPS handling

* Update training_args.py

* Cleanup

src/transformers/training_args.py

@@ -48,6 +48,7 @@
     is_torch_bf16_cpu_available,
     is_torch_bf16_gpu_available,
     is_torch_mlu_available,
+    is_torch_mps_available,
     is_torch_neuroncore_available,
     is_torch_npu_available,
     is_torch_tf32_available,
@@ -2178,6 +2179,8 @@ def _setup_devices(self) -> "torch.device":
                     )
             if self.use_cpu:
                 device = torch.device("cpu")
+            elif is_torch_mps_available():
+                device = torch.device("mps")
             elif is_torch_xpu_available():
                 if not is_ipex_available() and not is_accelerate_available("0.32.0.dev"):
                     raise ImportError("Using the XPU PyTorch backend requires `accelerate>=0.32.0.dev`")