NER is the task of labelling words in a sentence as persons, locations, etc.
The task, in its entirety, involves selecting a sub-label amongst an hierarchy of NER labels. This implementation, though, does not delve into this hierarchy, rather the labels are clubbed until 9 high-level NER tags are left. BIO encoding has been used.
A hybrid biLSTM-CRF model is used, much as outlined in this paper. A fast implementation of linear chain CRFs with fully vectorized training is provided.
The dataset: Publicly available GMB dataset, see https://gmb.let.rug.nl/data.php
More about the NER task can be read here