import tensorflow from .base import BaseTokenizer class PlainTokenizer(BaseTokenizer): """ Tokenizer which just splits the text into tokens by separating them at whitespaces. """ def tokenize_plain(self, text: str) -> str: return text def tokenize_tensorflow(self, text: tensorflow.Tensor) -> tensorflow.Tensor: return text