Hello NLPCrafters,

I would like to suggest adding support for Flair to complement the current 
token providers that we have.

Flair is an open source framework for NLP that is being developed. Flair is a 
simple framework for NLP and is newer than Spacy. Flair aims for 
state-of-the-art (SOTA) performance for natural language tasks. It is developed 
by Humboldt University of Berlin. I have been using Flair for certain tasks at 
my workplace and while I find that it is slower than Spacy, it performs 
extremely well in the tagging and sentiment analysis tasks that I have used it 
for, It is currently SOTA for several text analytics tasks and supports various 
types of word embeddings  (GloVe, BERT, FastText etc.).

Flair Github: https://github.com/flairNLP/flair

Side note: I also think that we should make the model options (be it Spacy or 
Flair) configurable and add them to the nlpcraft.conf file and let the 
developer change the model that needs to be downloaded. For example, for Spacy, 
it would be the small (en_core_web_sm), medium (en_core_web_md), large 
(en_core_web_lg) or transformer pipeline (en_core_web_trf) English models.

Let me know what you think


Thanks,

Rahul Padmanabhan

Reply via email to