1.79 kB
| --- | |
| language: | |
| - "ja" | |
| tags: | |
| - "japanese" | |
| - "token-classification" | |
| - "pos" | |
| - "wikipedia" | |
| - "dependency-parsing" | |
| datasets: | |
| - "universal_dependencies" | |
| license: "cc-by-sa-4.0" | |
| pipeline_tag: "token-classification" | |
| widget: | |
| - text: "国境の長いトンネルを抜けると雪国であった。" | |
| --- | |
| # bert-large-japanese-luw-upos | |
| ## Model Description | |
| This is a BERT model pre-trained on Japanese Wikipedia texts for POS-tagging and dependency-parsing, derived from [bert-large-japanese-char-extended](https://huggingface.co/KoichiYasuoka/bert-large-japanese-char-extended). Every long-unit-word is tagged by [UPOS](https://universaldependencies.org/u/pos/) (Universal Part-Of-Speech) and [FEATS](https://universaldependencies.org/u/feat/). | |
| ## How to Use | |
| ```py | |
| import torch | |
| from transformers import AutoTokenizer,AutoModelForTokenClassification | |
| tokenizer=AutoTokenizer.from_pretrained("KoichiYasuoka/bert-large-japanese-luw-upos") | |
| model=AutoModelForTokenClassification.from_pretrained("KoichiYasuoka/bert-large-japanese-luw-upos") | |
| s="国境の長いトンネルを抜けると雪国であった。" | |
| p=[model.config.id2label[q] for q in torch.argmax(model(tokenizer.encode(s,return_tensors="pt"))["logits"],dim=2)[0].tolist()[1:-1]] | |
| print(list(zip(s,p))) | |
| ``` | |
| or | |
| ```py | |
| import esupar | |
| nlp=esupar.load("KoichiYasuoka/bert-large-japanese-luw-upos") | |
| print(nlp("国境の長いトンネルを抜けると雪国であった。")) | |
| ``` | |
| ## Reference | |
| 安岡孝一: [Transformersと国語研長単位による日本語係り受け解析モデルの製作](http://id.nii.ac.jp/1001/00216223/), 情報処理学会研究報告, Vol.2022-CH-128, No.7 (2022年2月), pp.1-8. | |
| ## See Also | |
| [esupar](https://github.com/KoichiYasuoka/esupar): Tokenizer POS-tagger and Dependency-parser with BERT/RoBERTa models | |