Hi, I’m using a BucketIterator with a batch size of 64 to iterate over my training and test dataset.
train_iterator, test_iterator = BucketIterator.splits(
sort=True, sort_key=lambda x: len(x.prot_seq),
The datasets are not a multiple of 64 though.
When training, I’m getting the following error message:
If training, sequence Length 35 has to be a multiple of least common multiple chunk_length 64. Please consider padding the input to a length of 64
Shouldn’t the BucketIterator take care of always giving batches of 64 sequences?
Or is this somehow configurable? The docs are not so helpful in this regard.
Btw. I’m using a Reformer model
from transformers import ReformerConfig, ReformerModelWithLMHead
Thanks for your help in advance!