MultiTagger for Multiple Sequence Tagging Models

[alanakbik]

With the upcoming biomedical Flair release ("hunflair") comes a new object called MultiTagger with which multiple different SequenceTagger can be chained.

This tasks monitors the optimization of this model, namely:

• If two models use the same static embedding, only one of them should be kept in memory
• If two models use the same static embedding, only one of them should be applied during prediction

The idea is to enable us to load several models at once, like this:

# load tagger for POS, chunking, NER and frame detection
tagger = MultiTagger.load(['pos', 'upos', 'chunk', 'ner', 'frame'])

# example sentence
sentence = Sentence("George Washington was born in Washington")

# predict
tagger.predict(sentence)

[alanakbik]

Added in #1790

[shenron9155]

I have a custom flair LM model and a few classifiers that use the same LM embdeddings. Is there an equivalent MultiClassifier? So that the embeddings are reused per inference.