NLP Model Training Statistic Calculation

Is there a way to disable / why does Abbyy need to perform the task "Calculating statistics for current NLP moels in training batch" according to the training log file?

Almost half the NLP training batch processing time is being taken up with this statistic calculation task according to the log file.

This is moving the training of our NLP model from 4.5 hours to 8.5.

Is this needed every time we run a NLP model training batch?

If it can not be disabled, are there any other ways of improving NLP model training, as it is running on a 16vCore Azure VM and continuing to take 8+ hours each training batch.




Please sign in to leave a comment.