Aymara

This is a partial release of Deeplima models, including segmentation and tagging models, as well as a few lemmatization models.
These models have been trained with a multi-objective optimization strategy. The best ones for tokens, upos and lemmas have been selected.

v0.1.6-beta-morph


v0.1.6-beta-en-c4medium

Tokenization only (without sentence segmentation) models trained on all trainable corpora from UD 2.8 collection (67 languages), 8 models for each corpus.
Deeplima commit: https://github.com/aymara/lima/commit/40ab9ad6e53fde56daf8b357ea0bc078d0545968
LSTM hidden state: 8

- lemmatization models updated
- speed improved
- Albanian language added

Starting from this version models are packaged per language.

Models are installed to /usr/.