ai-agents-mps-support-trainingarguments-does-not-automatically-detect-mp-bf393b27

THEN Subclass TrainingArguments and override the `device` property to check `torch.backends.mps.is_available()`. If available, return `torch.device('mps')` before falling back to CPU. This forces the Trainer to use the MPS GPU.

import torch
from transformers import TrainingArguments

class TrainingArgumentsWithMPSSupport(TrainingArguments):
    @property
    def device(self) -> torch.device:
        if torch.cuda.is_available():
            return torch.device("cuda")
        elif torch.backends.mps.is_available():
            return torch.device("mps")
        else:
            return torch.device("cpu")