OFA icon indicating copy to clipboard operation
OFA copied to clipboard

unify_task.py is not completed

Open flymark2010 opened this issue 2 years ago • 1 comments

fairseq.data.FairseqDataset, fairseq.data.iterators, math are missed in file unify_task.py.

I use pretrain_data_examples.zip you have provided, follow the description pretraining, and run

cd run_scripts/pretraining
bash pretrain_ofa_base.sh

Error got as follows:

  ...
  File "/workspace/project/OFA/trainer.py", line 664, in get_train_iterator
    disable_iterator_cache=disable_iterator_cache,
  File "/workspace/project/OFA/tasks/pretrain_tasks/unify_task.py", line 172, in get_batch_iterator
    assert isinstance(dataset, FairseqDataset)
NameError: name 'FairseqDataset' is not defined
  ...
  File "/workspace/project/OFA/trainer.py", line 664, in get_train_iterator
    disable_iterator_cache=disable_iterator_cache,
  File "/workspace/project/OFA/tasks/pretrain_tasks/unify_task.py", line 184, in get_batch_iterator
    num_batches = math.ceil(math.ceil(total_row_count / num_shards) / max_sentences)
NameError: name 'math' is not defined
  ...
  File "/workspace/project/OFA/trainer.py", line 664, in get_train_iterator
    disable_iterator_cache=disable_iterator_cache,
  File "/workspace/project/OFA/tasks/pretrain_tasks/unify_task.py", line 190, in get_batch_iterator
    epoch_iter = iterators.EpochBatchIterator(
NameError: name 'iterators' is not defined

flymark2010 avatar Jul 25 '22 03:07 flymark2010

@flymark2010 Thanks for your feedback! We forgot to import the corresponding packages in the last update. The latest commit have imported the missing packages, please check out our latest code and have a try.

logicwong avatar Jul 25 '22 07:07 logicwong