About Loss Value

[jnanliu]

Thanks for the great package. In my work, I use the -log_likehood score as the loss, however, when training after several batches, the value of -log_likehood becomes negative. Is this normal?