models.md

Models

COMBO provides pre-trained models for:

• morphosyntactic prediction (i.e. part-of-speech tagging, morphosyntactic analysis, lemmatisation and dependency parsing) trained on the treebanks from Universal Dependencies repository (Zeman et al. 2020),
• enhanced dependency parsing trained on IWPT 2020 shared task data (Bouma et al. 2020).

Pre-trained models

Morphosyntactic prediction models trained on the selected UD treebanks version 2.7 and their evaluation results are listed in Model performance (UD2.7) table.

Morphosyntactic prediction models trained on the seleted UD treebanks version 2.5 and enhanced parsing models are listed in the spreadsheets: UD2.5-trained COMBO models and enhanced COMBO models.

License

Models are distributed under the same license as datasets used for their training.

See Universal Dependencies v2.7 License Agreement and Universal Dependencies v2.5 License Agreement for details.

Automatic download

The pre-trained models can be automatically downloaded with the from_pretrained method in the Python mode. Select the model name of a pre-trained model (see the column Model name in Model performance (UD2.7), UD2.5-trained COMBO models and enhanced COMBO models) and pass the name as an attribute of the from_pretrained method:

from combo.predict import COMBO

nlp = COMBO.from_pretrained("polish-herbert-base")

If the model name doesn't match any model on the pre-trained model lists, COMBO looks for a model in local env.

Manual download

If you want to use COMBO in the command-line mode, you need to manually download a pre-trained model. The pre-trained models can be manually downloaded to a local disk with the wget package. The links to the pre-trained models are listed in the column Model name in Model performance (UD2.7), or Model link in UD2.5-trained COMBO models and enhanced COMBO models.

wget http://mozart.ipipan.waw.pl/~mklimaszewski/models/polish-herbert-base.tar.gz

The path to the downloaded model should be passed as a parameter for COMBO in CLI (see prediction doc).