chunked/nltk_chunking/_perf_train_special/chunk_9.txt

pip install torch torchvision torchaudio
[TrainingArguments] uses the mps device by default if it's available which means you don't need to explicitly set the device.