Data: wtorek 12.01.2021, godz. 11:00-12:00
Prelegent: Maxime Méloux (Samsung/UAM)
Abstrakt: I will present various methods of compression for part-of-speech tagger models, which allow to reduce on-disk model size and increase inference speed while preserving model performance as best as possible. I will particularly focus on compression methods for static and dynamic embeddings, such as Flair embeddings (https://www.aclweb.org/anthology/C18-1139).