from transformers import BertTokenizerFast
from capreolus import ConfigOption
from . import Tokenizer
[docs]@Tokenizer.register
class BertTokenizer(Tokenizer):
[docs] module_name = "berttokenizer"
[docs] config_spec = [ConfigOption("pretrained", "bert-base-uncased", "pretrained model to load vocab from")]
[docs] def build(self):
self.bert_tokenizer = BertTokenizerFast.from_pretrained(self.config["pretrained"])
[docs] def convert_tokens_to_ids(self, tokens):
return self.bert_tokenizer.convert_tokens_to_ids(tokens)
[docs] def tokenize(self, sentences):
if not sentences or len(sentences) == 0: # either "" or []
return []
if isinstance(sentences, str):
return self.bert_tokenizer.tokenize(sentences)
return [self.bert_tokenizer.tokenize(s) for s in sentences]