assert gnorm >= 0, 'encounter nan in gradients

[mustfkeskin]

Hello

I get the following error when I try to train the model with my own training data.
What could be the error

[hitvoice]

Can you get the batch which causes NaN? Maybe by catching the exception and save the batch data (feed_dict).

[mustfkeskin]

I got 1 empty string "" in my training data.
I drop it then it works

Thanks