Nans found in input

[Reham-Osama]

When I try to train the baseline the following warning is logged many times:
"Warning: NaN or Inf found in input tensor."
Note that it happens only during training, doesn't happen during validation.

Is there something wrong in the baseline configurations or code that causes this issue to appear so frequently?

[alontalmor]

Hi can you please send the command you used when this happened ? thanks

[Reham-Osama]

This is the command I used
"python -m allennlp.run train baseline/MRQA_BERTbase.jsonnet -s output_dir --include-package mrqa_allennlp"