Views
No views yet
FacebookAI/xlm-roberta-base as its base encoder and is trained on UD_Serbian-SET (UD v2.17).| Metric | Tokens | Words | Sentences |
|---|---|---|---|
| F1 | 100.00 | 100.00 | 97.78 |
pip install combo-seg1from combo_seg import ComboSeg
2
3# Load a pre-trained model
4nlp = ComboSeg("Serbian")
5
6# Segment raw text — returns Document with hierarchy: Document -> Turn -> Sentence -> Token
7doc = nlp("Брза смеђа лисица прескаче лењог пса.")
8
9# Inspect results
10for turn in doc.turns:
11 for sentence in turn.sentences:
12 print(f"Sentence: {sentence.text}")
13 for token in sentence.tokens:
14 if token.is_multi_word:
15 print(f" MWT: {token.text} -> {token.subwords}")
16 else:
17 print(f" Token: {token.text}")1from combo_seg import ComboSeg
2
3nlp = ComboSeg.from_pretrained("clarin-pl/combo-seg-xlm-roberta-base-serbian-set-ud2.17")
4doc = nlp("Брза смеђа лисица прескаче лењог пса.")LICENSE.txt file in the treebank repository:1@software{combo_seg_2026,
2 author = {Ulewicz, Michał and Wróblewska, Alina},
3 title = {{COMBO-SEG} Models Trained on {UD} v2.17},
4 year = {2026},
5 publisher = {Zenodo},
6 doi = {10.5281/zenodo.19651441},
7 url = {https://doi.org/10.5281/zenodo.19651441}
8}