Humboldt-Universität zu Berlin - Faculty of Mathematics and Natural Sciences - Machine Learning

The Flair framework is our open source framework for state-of-the-art NLP, built on our group's machine learning research. It is freely available and already used in hundeds of research projects and industrial applications. As official part of the PyTorch ecosystem, Flair is one of the most popular deep learning frameworks for NLP.

Flair is currently state-of-the-art across a range of text analytics tasks for text data in many different languages such as German, English, Polish, Japanese, etc. It is being developed by our group, in collaboration with the open source community and industrial partners like Zalando Resarch. If you're interested in using Flair, do contact us or help us develop it by joining our open source community.

Analyzing Text Data

Flair supports many types of text analsis, including sequence labeling, text classification, similarity learning and text regression. For instance, Flair is commonly used for Named Entity Recognition (NER), as shown in this use case from our partner Zalando SE:

Word Embeddings

Flair supports a large number of popular and experimental word embeddings for application to NLP tasks, including GloVe, FastText, ELMo, BERT and its variants (RoBERTa, ALBERT, CamemBERT, etc.), XLM, Byte Pair Embeddings and of course our own Flair embeddings:

Our framework hides the complexity of each embedding behind a unified interface, making it easy to "mix and match" embeddings for your use case. This for instance allows users to compare approaches such as BERT to other approaches and leverage them in their models. 

More Information