test_tokenization_electra.py

# coding=utf-8
# Copyright 2023 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import os
import unittest

from transformers import ElectraTokenizerFast
from transformers.models.electra.tokenization_electra import (
    VOCAB_FILES_NAMES,
    BasicTokenizer,
    ElectraTokenizer,
    WordpieceTokenizer,
    _is_control,
    _is_punctuation,
    _is_whitespace,
)
from transformers.testing_utils import require_tokenizers, slow

from ...test_tokenization_common import TokenizerTesterMixin, filter_non_english


@require_tokenizers
class ElectraTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
    from_pretrained_id = "google/electra-small-generator"
    tokenizer_class = ElectraTokenizer
    rust_tokenizer_class = ElectraTokenizerFast
    test_rust_tokenizer = True
    space_between_special_tokens = True
    from_pretrained_filter = filter_non_english

    def setUp(self):
        super().setUp()

        vocab_tokens = [
            "[UNK]",
            "[CLS]",
            "[SEP]",
            "[PAD]",
            "[MASK]",
            "want",
            "##want",
            "##ed",
            "wa",
            "un",
            "runn",
            "##ing",
            ",",
            "low",
            "lowest",
        ]
        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
        with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
            vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))

    def get_input_output_texts(self, tokenizer):
        input_text = "UNwant\u00e9d,running"
        output_text = "unwanted, running"
        return input_text, output_text

    def test_full_tokenizer(self):
        tokenizer = self.tokenizer_class(self.vocab_file)

        tokens = tokenizer.tokenize("UNwant\u00e9d,running")
        self.assertListEqual(tokens, ["un", "##want", "##ed", ",", "runn", "##ing"])
        self.assertListEqual(tokenizer.convert_tokens_to_ids(tokens), [9, 6, 7, 12, 10, 11])

    def test_rust_and_python_full_tokenizers(self):
        if not self.test_rust_tokenizer:
            return

        tokenizer = self.get_tokenizer()
        rust_tokenizer = self.get_rust_tokenizer()

        sequence = "UNwant\u00e9d,running"

        tokens = tokenizer.tokenize(sequence)
        rust_tokens = rust_tokenizer.tokenize(sequence)
        self.assertListEqual(tokens, rust_tokens)

        ids = tokenizer.encode(sequence, add_special_tokens=False)
        rust_ids = rust_tokenizer.encode(sequence, add_special_tokens=False)
        self.assertListEqual(ids, rust_ids)

        rust_tokenizer = self.get_rust_tokenizer()
        ids = tokenizer.encode(sequence)
        rust_ids = rust_tokenizer.encode(sequence)
        self.assertListEqual(ids, rust_ids)

        # With lower casing
        tokenizer = self.get_tokenizer(do_lower_case=True)
        rust_tokenizer = self.get_rust_tokenizer(do_lower_case=True)

        sequence = "UNwant\u00e9d,running"

        tokens = tokenizer.tokenize(sequence)
        rust_tokens = rust_tokenizer.tokenize(sequence)
        self.assertListEqual(tokens, rust_tokens)

        ids = tokenizer.encode(sequence, add_special_tokens=False)
        rust_ids = rust_tokenizer.encode(sequence, add_special_tokens=False)
        self.assertListEqual(ids, rust_ids)

        rust_tokenizer = self.get_rust_tokenizer()
        ids = tokenizer.encode(sequence)
        rust_ids = rust_tokenizer.encode(sequence)
        self.assertListEqual(ids, rust_ids)

    def test_chinese(self):
        tokenizer = BasicTokenizer()

        self.assertListEqual(tokenizer.tokenize("ah\u535a\u63a8zz"), ["ah", "\u535a", "\u63a8", "zz"])

    def test_basic_tokenizer_lower(self):
        tokenizer = BasicTokenizer(do_lower_case=True)

        self.assertListEqual(
            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU?  "), ["hello", "!", "how", "are", "you", "?"]
        )
        self.assertListEqual(tokenizer.tokenize("H\u00e9llo"), ["hello"])

    def test_basic_tokenizer_lower_strip_accents_false(self):
        tokenizer = BasicTokenizer(do_lower_case=True, strip_accents=False)

        self.assertListEqual(
            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hällo", "!", "how", "are", "you", "?"]
        )
        self.assertListEqual(tokenizer.tokenize("H\u00e9llo"), ["h\u00e9llo"])

    def test_basic_tokenizer_lower_strip_accents_true(self):
        tokenizer = BasicTokenizer(do_lower_case=True, strip_accents=True)

        self.assertListEqual(
            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hallo", "!", "how", "are", "you", "?"]
        )
        self.assertListEqual(tokenizer.tokenize("H\u00e9llo"), ["hello"])

    def test_basic_tokenizer_lower_strip_accents_default(self):
        tokenizer = BasicTokenizer(do_lower_case=True)

        self.assertListEqual(
            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hallo", "!", "how", "are", "you", "?"]
        )
        self.assertListEqual(tokenizer.tokenize("H\u00e9llo"), ["hello"])

    def test_basic_tokenizer_no_lower(self):
        tokenizer = BasicTokenizer(do_lower_case=False)

        self.assertListEqual(
            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU?  "), ["HeLLo", "!", "how", "Are", "yoU", "?"]
        )

    def test_basic_tokenizer_no_lower_strip_accents_false(self):
        tokenizer = BasicTokenizer(do_lower_case=False, strip_accents=False)

        self.assertListEqual(
            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["HäLLo", "!", "how", "Are", "yoU", "?"]
        )

    def test_basic_tokenizer_no_lower_strip_accents_true(self):
        tokenizer = BasicTokenizer(do_lower_case=False, strip_accents=True)

        self.assertListEqual(
            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["HaLLo", "!", "how", "Are", "yoU", "?"]
        )

    def test_basic_tokenizer_respects_never_split_tokens(self):
        tokenizer = BasicTokenizer(do_lower_case=False, never_split=["[UNK]"])

        self.assertListEqual(
            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU? [UNK]"), ["HeLLo", "!", "how", "Are", "yoU", "?", "[UNK]"]
        )

    def test_wordpiece_tokenizer(self):
        vocab_tokens = ["[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn", "##ing"]

        vocab = {}
        for i, token in enumerate(vocab_tokens):
            vocab[token] = i
        tokenizer = WordpieceTokenizer(vocab=vocab, unk_token="[UNK]")

        self.assertListEqual(tokenizer.tokenize(""), [])

        self.assertListEqual(tokenizer.tokenize("unwanted running"), ["un", "##want", "##ed", "runn", "##ing"])

        self.assertListEqual(tokenizer.tokenize("unwantedX running"), ["[UNK]", "runn", "##ing"])

    def test_is_whitespace(self):
        self.assertTrue(_is_whitespace(" "))
        self.assertTrue(_is_whitespace("\t"))
        self.assertTrue(_is_whitespace("\r"))
        self.assertTrue(_is_whitespace("\n"))
        self.assertTrue(_is_whitespace("\u00a0"))

        self.assertFalse(_is_whitespace("A"))
        self.assertFalse(_is_whitespace("-"))

    def test_is_control(self):
        self.assertTrue(_is_control("\u0005"))

        self.assertFalse(_is_control("A"))
        self.assertFalse(_is_control(" "))
        self.assertFalse(_is_control("\t"))
        self.assertFalse(_is_control("\r"))

    def test_is_punctuation(self):
        self.assertTrue(_is_punctuation("-"))
        self.assertTrue(_is_punctuation("$"))
        self.assertTrue(_is_punctuation("`"))
        self.assertTrue(_is_punctuation("."))

        self.assertFalse(_is_punctuation("A"))
        self.assertFalse(_is_punctuation(" "))

    def test_clean_text(self):
        tokenizer = self.get_tokenizer()
        rust_tokenizer = self.get_rust_tokenizer()

        # Example taken from the issue https://github.com/huggingface/tokenizers/issues/340
        self.assertListEqual([tokenizer.tokenize(t) for t in ["Test", "\xad", "test"]], [["[UNK]"], [], ["[UNK]"]])

        self.assertListEqual(
            [rust_tokenizer.tokenize(t) for t in ["Test", "\xad", "test"]], [["[UNK]"], [], ["[UNK]"]]
        )

    @slow
    def test_sequence_builders(self):
        tokenizer = self.tokenizer_class.from_pretrained("google/electra-base-discriminator")

        text = tokenizer.encode("sequence builders", add_special_tokens=False)
        text_2 = tokenizer.encode("multi-sequence build", add_special_tokens=False)

        encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
        encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)

        assert encoded_sentence == [101] + text + [102]
        assert encoded_pair == [101] + text + [102] + text_2 + [102]

    def test_offsets_with_special_characters(self):
        for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
            with self.subTest(f"{tokenizer.__class__.__name__} ({pretrained_name})"):
                tokenizer_r = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs)

                sentence = f"A, naïve {tokenizer_r.mask_token} AllenNLP sentence."
                tokens = tokenizer_r.encode_plus(
                    sentence,
                    return_attention_mask=False,
                    return_token_type_ids=False,
                    return_offsets_mapping=True,
                    add_special_tokens=True,
                )

                do_lower_case = tokenizer_r.do_lower_case if hasattr(tokenizer_r, "do_lower_case") else False
                expected_results = (
                    [
                        ((0, 0), tokenizer_r.cls_token),
                        ((0, 1), "A"),
                        ((1, 2), ","),
                        ((3, 5), "na"),
                        ((5, 6), "##ï"),
                        ((6, 8), "##ve"),
                        ((9, 15), tokenizer_r.mask_token),
                        ((16, 21), "Allen"),
                        ((21, 23), "##NL"),
                        ((23, 24), "##P"),
                        ((25, 33), "sentence"),
                        ((33, 34), "."),
                        ((0, 0), tokenizer_r.sep_token),
                    ]
                    if not do_lower_case
                    else [
                        ((0, 0), tokenizer_r.cls_token),
                        ((0, 1), "a"),
                        ((1, 2), ","),
                        ((3, 8), "naive"),
                        ((9, 15), tokenizer_r.mask_token),
                        ((16, 21), "allen"),
                        ((21, 23), "##nl"),
                        ((23, 24), "##p"),
                        ((25, 33), "sentence"),
                        ((33, 34), "."),
                        ((0, 0), tokenizer_r.sep_token),
                    ]
                )

                self.assertEqual(
                    [e[1] for e in expected_results], tokenizer_r.convert_ids_to_tokens(tokens["input_ids"])
                )
                self.assertEqual([e[0] for e in expected_results], tokens["offset_mapping"])

    def test_change_tokenize_chinese_chars(self):
        list_of_commun_chinese_char = ["的", "人", "有"]
        text_with_chinese_char = "".join(list_of_commun_chinese_char)
        for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
            with self.subTest(f"{tokenizer.__class__.__name__} ({pretrained_name})"):
                kwargs["tokenize_chinese_chars"] = True
                tokenizer_p = self.tokenizer_class.from_pretrained(pretrained_name, **kwargs)
                tokenizer_r = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs)

                ids_without_spe_char_p = tokenizer_p.encode(text_with_chinese_char, add_special_tokens=False)
                ids_without_spe_char_r = tokenizer_r.encode(text_with_chinese_char, add_special_tokens=False)

                tokens_without_spe_char_r = tokenizer_r.convert_ids_to_tokens(ids_without_spe_char_r)
                tokens_without_spe_char_p = tokenizer_p.convert_ids_to_tokens(ids_without_spe_char_p)

                # it is expected that each Chinese character is not preceded by "##"
                self.assertListEqual(tokens_without_spe_char_p, list_of_commun_chinese_char)
                self.assertListEqual(tokens_without_spe_char_r, list_of_commun_chinese_char)

                kwargs["tokenize_chinese_chars"] = False
                tokenizer_r = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs)
                tokenizer_p = self.tokenizer_class.from_pretrained(pretrained_name, **kwargs)

                ids_without_spe_char_r = tokenizer_r.encode(text_with_chinese_char, add_special_tokens=False)
                ids_without_spe_char_p = tokenizer_p.encode(text_with_chinese_char, add_special_tokens=False)

                tokens_without_spe_char_r = tokenizer_r.convert_ids_to_tokens(ids_without_spe_char_r)
                tokens_without_spe_char_p = tokenizer_p.convert_ids_to_tokens(ids_without_spe_char_p)

                # it is expected that only the first Chinese character is not preceded by "##".
                expected_tokens = [
                    f"##{token}" if idx != 0 else token for idx, token in enumerate(list_of_commun_chinese_char)
                ]
                self.assertListEqual(tokens_without_spe_char_p, expected_tokens)
                self.assertListEqual(tokens_without_spe_char_r, expected_tokens)