def test_tokens_to_text(self, test_data_dir):
tokenizer = SentencePieceTokenizer(test_data_dir + self.model_name)
text = "[CLS] a b c [MASK] e f [SEP] g h i [SEP]"
tokens = tokenizer.text_to_tokens(text)
result = tokenizer.tokens_to_text(tokens)
assert text == result
def test_tokens_to_text(self, test_data_dir):
tokenizer = SentencePieceTokenizer(test_data_dir + self.model_name)
# <cls> is user_defined_symbol in the test tokenizer model
text = "<cls> a b c e f g h i"
tokens = tokenizer.text_to_tokens(text)
result = tokenizer.tokens_to_text(tokens)
assert text == result
def test_text_to_tokens(self, test_data_dir):
tokenizer = SentencePieceTokenizer(test_data_dir + self.model_name)
# <cls> is user_defined_symbol in the test tokenizer model
# <unk>, <sep>, <s>, and </s> are control symbols
text = "<cls> a b c <sep> e f g h i </s>"
tokens = tokenizer.text_to_tokens(text)
assert tokens.count("<cls>") == 1
assert tokens.count("<sep>") == 0
assert tokens.count("</s>") == 0
def test_text_to_tokens(self, test_data_dir):
tokenizer = SentencePieceTokenizer(test_data_dir + self.model_name)
special_tokens = MODEL_SPECIAL_TOKENS
tokenizer.add_special_tokens(special_tokens)
text = "[CLS] a b c [MASK] e f [SEP] g h i [SEP]"
tokens = tokenizer.text_to_tokens(text)
assert len(tokens) == len(text.split())
assert tokens.count("[CLS]") == 1
assert tokens.count("[MASK]") == 1
assert tokens.count("[SEP]") == 2
def test_ids_to_tokens(self, test_data_dir):
tokenizer = SentencePieceTokenizer(test_data_dir + self.model_name)
special_tokens = MODEL_SPECIAL_TOKENS
tokenizer.add_special_tokens(special_tokens)
text = "[CLS] a b c [MASK] e f [SEP] g h i [SEP]"
tokens = tokenizer.text_to_tokens(text)
ids = tokenizer.tokens_to_ids(tokens)
result = tokenizer.ids_to_tokens(ids)
assert len(result) == len(tokens)
for i in range(len(result)):
assert result[i] == tokens[i]
def test_tokens_to_ids(self, test_data_dir):
tokenizer = SentencePieceTokenizer(test_data_dir + self.model_name,
legacy=True)
special_tokens = MODEL_SPECIAL_TOKENS
tokenizer.add_special_tokens(special_tokens)
text = "[CLS] a b c [MASK] e f [SEP] g h i [SEP]"
tokens = tokenizer.text_to_tokens(text)
ids = tokenizer.tokens_to_ids(tokens)
assert len(ids) == len(tokens)
assert ids.count(tokenizer.token_to_id("[CLS]")) == 1
assert ids.count(tokenizer.token_to_id("[MASK]")) == 1
assert ids.count(tokenizer.token_to_id("[SEP]")) == 2
class EnJaTokenizer:
"""
Tokenizer for Japanese & English that does Moses tokenization followed by SentencePiece
Args:
sp_tokenizer_model_path: String path to a sentencepiece model
lang_id: One of ['en', 'ja'].
"""
def __init__(self, sp_tokenizer_model_path: str, lang_id: str):
self.moses_tokenizer = MosesTokenizer(lang=lang_id)
self.sp_tokenizer = SentencePieceTokenizer(model_path=sp_tokenizer_model_path)
def sp_tokenize(self, text: str) -> str:
return ' '.join(self.sp_tokenizer.text_to_tokens(text))
def tokenize(self, text, escape=False, return_str=False):
"""
Tokenizes text using Moses -> Sentencepiece.
"""
text = self.moses_tokenizer.tokenize(text, escape=escape, return_str=True)
text = self.sp_tokenize(text)
return text if return_str else text.split()