Num_docs and evaluating docs

[linguist89]

Hey everyone, I have two questions:

1. I'm curious how to determine the num_docs hypterparameter? The default setting is 2802 docs, but I'm using my own data and the data is 1 long training file that is comprised of 72 "#begin" and "#end" lines in it's CoNLL format. This could equate to 72 separate lines when converted to the jsonlines format. Is this the correct way to determine the doc_num?

2. When evaluating it says evaluating x/420 (x is just the lines it's evaluating). However, how is this 420 determined? According to the explanation in my previous question, my test set would only be 11 documents and I don't have any datasets that are 420. Any ideas about this value?

Thank you!