NameError: name 'AcceleratorState' is not defined

[ezyang]

System Info

This script fails on 68d53bc but passes on de9255d

Hopefully the problem is pretty clear from the message.

(/home/ezyang/local/debug/pytorch-env) [ezyang@devgpu019.ftw1 ~/local/debug]$ pp python transformers/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py     --dataset_name="common_voice"     --model_name_or_path="facebook/wav2vec2-large-xlsr-53"     --dataset_config_name="tr"     --output_dir="./wav2vec2-common_voice-tr-demo-dist"     --preprocessing_num_workers="16"     --overwrite_output_dir     --num_train_epochs="15"     --per_device_train_batch_size="4"     --gradient_accumulation_steps="1"     --learning_rate="3e-4"     --warmup_steps="500"     --evaluation_strategy="steps"     --text_column_name="sentence"     --save_steps="400"     --eval_steps="100"     --logging_steps="1"     --layerdrop="0.0"     --save_total_limit="3"     --freeze_feature_encoder     --gradient_checkpointing     --chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” �     --fp16     --group_by_length     --do_train --do_eval     --torch_compile True
Traceback (most recent call last):
  File "/data/users/ezyang/debug/transformers/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py", line 775, in <module>
    main()
  File "/data/users/ezyang/debug/transformers/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py", line 380, in main
    model_args, data_args, training_args = parser.parse_args_into_dataclasses()
  File "/home/ezyang/local/debug/pytorch-env/lib/python3.10/site-packages/transformers/hf_argparser.py", line 346, in parse_args_into_dataclasses
    obj = dtype(**inputs)
  File "<string>", line 111, in __init__
  File "/home/ezyang/local/debug/pytorch-env/lib/python3.10/site-packages/transformers/training_args.py", line 1340, in __post_init__
    and (self.device.type != "cuda")
  File "/home/ezyang/local/debug/pytorch-env/lib/python3.10/site-packages/transformers/training_args.py", line 1764, in device
    return self._setup_devices
  File "/home/ezyang/local/debug/pytorch-env/lib/python3.10/site-packages/transformers/utils/generic.py", line 54, in __get__
    cached = self.fget(obj)
  File "/home/ezyang/local/debug/pytorch-env/lib/python3.10/site-packages/transformers/training_args.py", line 1670, in _setup_devices
    AcceleratorState._reset_state()
NameError: name 'AcceleratorState' is not defined

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

python transformers/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py --dataset_name="common_voice" --model_name_or_path="facebook/wav2vec2-large-xlsr-53" --dataset_config_name="tr" --output_dir="./wav2vec2-common_voice-tr-demo-dist" --preprocessing_num_workers="16" --overwrite_output_dir --num_train_epochs="15" --per_device_train_batch_size="4" --gradient_accumulation_steps="1" --learning_rate="3e-4" --warmup_steps="500" --evaluation_strategy="steps" --text_column_name="sentence" --save_steps="400" --eval_steps="100" --logging_steps="1" --layerdrop="0.0" --save_total_limit="3" --freeze_feature_encoder --gradient_checkpointing --chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � --fp16 --group_by_length --do_train --do_eval --torch_compile True

Expected behavior

doesn't error this way

[sgugger]

I think you don't have accelerate installed, which is now a required dependency for the Trainer: pip install accelerate.

[ezyang]

Ah ok, in that case, probably a requirements.txt just needs to get updated somewhere. (Also, naively, I would have expected an ImportError if I had a missing dependency, not a NameError)

[muellerzr]

Thanks @ezyang, with #23906 we'll raise an ImportError properly :)