wyu97 / easy-use-bart Goto Github PK

It seems that the "k_out" parameter is not defined in the "DataTrainingArguments".

Could you please check it, and what is the meaning of "k_out"?

Thanks!

Traceback (most recent call last):
File "finetune.py", line 302, in
main()
File "finetune.py", line 270, in main
metrics.update(eval_top1_acc(out_pred_path, out_pred_ref, data_args.k_out)) ## top1_metrics
AttributeError: 'DataTrainingArguments' object has no attribute 'k_out'

Hi, thank you for your code, I try to run your code and I got an error about the hf_argparser.py file.
I try to search around but I don't understand why. (I guess it is maybe because the version that I am using is different with you?)
I'm using:

• Python 3.8.10 (this is different)
• transformers==3.3.1
• torch==1.7.0

If you have any clues about the error, please let me know.
Thank you very much for your support.

Here are the details about the error.

  File "finetune.py", line 302, in <module>
    main()
  File "finetune.py", line 103, in main
    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, Seq2SeqTrainingArguments))
  File "/home/xanh_2/vir_bart/lib/python3.8/site-packages/transformers/hf_argparser.py", line 40, in __init__
    self._add_dataclass_arguments(dtype)
  File "/home/xanh_2/vir_bart/lib/python3.8/site-packages/transformers/hf_argparser.py", line 80, in _add_dataclass_arguments
    elif hasattr(field.type, "__origin__") and issubclass(field.type.__origin__, List):
  File "/usr/lib/python3.8/typing.py", line 774, in __subclasscheck__
    return issubclass(cls, self.__origin__)
TypeError: issubclass() arg 1 must be a class

Hello, thank you for sharing nice job~ i want to reprdocue this project and then run the permGen project.
But when i run CUDA_VISIBLE_DEVICES=0 bash scripts/train_rocstory.sh, it always reminds:
RuntimeError: CUDA out of memory. Tried to allocate 982.00 MiB (GPU 0; 10.76 GiB total capacity; 7.80 GiB already allocated; 603.44 MiB free; 9.16 GiB reserved in total by PyTorch)
Prehaps it is the problem of batch_size, and i modify the line 363 in seq2seq_trainer.py with:
logger.info(" Total train batch size (w. parallel, distributed & accumulation) = %d", 64)
and per_device_train_batch_size=16 ,per_device_eval_batch_size=8 in train_rocstory.sh
It finally works, and i want to know Is there any other better way to modify the RunTimeError?