169 lines
7.6 KiB
Python
169 lines
7.6 KiB
Python
# coding=utf-8
|
|
# Copyright 2019 Hugging Face inc.
|
|
#
|
|
# 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 json
|
|
import os
|
|
import unittest
|
|
|
|
from transformers import DebertaTokenizer, DebertaTokenizerFast
|
|
from transformers.models.deberta.tokenization_deberta import VOCAB_FILES_NAMES
|
|
from transformers.testing_utils import slow
|
|
|
|
from ...test_tokenization_common import TokenizerTesterMixin
|
|
|
|
|
|
class DebertaTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
|
from_pretrained_id = "microsoft/deberta-base"
|
|
tokenizer_class = DebertaTokenizer
|
|
test_rust_tokenizer = True
|
|
rust_tokenizer_class = DebertaTokenizerFast
|
|
|
|
def setUp(self):
|
|
super().setUp()
|
|
|
|
# Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt
|
|
vocab = [
|
|
"l",
|
|
"o",
|
|
"w",
|
|
"e",
|
|
"r",
|
|
"s",
|
|
"t",
|
|
"i",
|
|
"d",
|
|
"n",
|
|
"\u0120",
|
|
"\u0120l",
|
|
"\u0120n",
|
|
"\u0120lo",
|
|
"\u0120low",
|
|
"er",
|
|
"\u0120lowest",
|
|
"\u0120newer",
|
|
"\u0120wider",
|
|
"[UNK]",
|
|
]
|
|
vocab_tokens = dict(zip(vocab, range(len(vocab))))
|
|
merges = ["#version: 0.2", "\u0120 l", "\u0120l o", "\u0120lo w", "e r", ""]
|
|
self.special_tokens_map = {"unk_token": "[UNK]"}
|
|
|
|
self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
|
|
self.merges_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["merges_file"])
|
|
with open(self.vocab_file, "w", encoding="utf-8") as fp:
|
|
fp.write(json.dumps(vocab_tokens) + "\n")
|
|
with open(self.merges_file, "w", encoding="utf-8") as fp:
|
|
fp.write("\n".join(merges))
|
|
|
|
def get_tokenizer(self, **kwargs):
|
|
kwargs.update(self.special_tokens_map)
|
|
return self.tokenizer_class.from_pretrained(self.tmpdirname, **kwargs)
|
|
|
|
def get_input_output_texts(self, tokenizer):
|
|
input_text = "lower newer"
|
|
output_text = "lower newer"
|
|
return input_text, output_text
|
|
|
|
def test_full_tokenizer(self):
|
|
tokenizer = self.get_tokenizer()
|
|
text = "lower newer"
|
|
bpe_tokens = ["l", "o", "w", "er", "\u0120", "n", "e", "w", "er"]
|
|
tokens = tokenizer.tokenize(text)
|
|
self.assertListEqual(tokens, bpe_tokens)
|
|
|
|
input_tokens = tokens + [tokenizer.unk_token]
|
|
input_bpe_tokens = [0, 1, 2, 15, 10, 9, 3, 2, 15, 19]
|
|
self.assertListEqual(tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
|
|
|
|
def test_token_type_ids(self):
|
|
tokenizer = self.get_tokenizer()
|
|
tokd = tokenizer("Hello", "World")
|
|
expected_token_type_ids = [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1]
|
|
self.assertListEqual(tokd["token_type_ids"], expected_token_type_ids)
|
|
|
|
@slow
|
|
def test_sequence_builders(self):
|
|
tokenizer = self.tokenizer_class.from_pretrained("microsoft/deberta-base")
|
|
|
|
text = tokenizer.encode("sequence builders", add_special_tokens=False)
|
|
text_2 = tokenizer.encode("multi-sequence build", add_special_tokens=False)
|
|
|
|
encoded_text_from_decode = tokenizer.encode(
|
|
"sequence builders", add_special_tokens=True, add_prefix_space=False
|
|
)
|
|
encoded_pair_from_decode = tokenizer.encode(
|
|
"sequence builders", "multi-sequence build", add_special_tokens=True, add_prefix_space=False
|
|
)
|
|
|
|
encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
|
|
encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)
|
|
|
|
assert encoded_sentence == encoded_text_from_decode
|
|
assert encoded_pair == encoded_pair_from_decode
|
|
|
|
@slow
|
|
def test_tokenizer_integration(self):
|
|
tokenizer_classes = [self.tokenizer_class]
|
|
if self.test_rust_tokenizer:
|
|
tokenizer_classes.append(self.rust_tokenizer_class)
|
|
|
|
for tokenizer_class in tokenizer_classes:
|
|
tokenizer = tokenizer_class.from_pretrained("microsoft/deberta-base")
|
|
|
|
sequences = [
|
|
"ALBERT: A Lite BERT for Self-supervised Learning of Language Representations",
|
|
"ALBERT incorporates two parameter reduction techniques",
|
|
"The first one is a factorized embedding parameterization. By decomposing the large vocabulary"
|
|
" embedding matrix into two small matrices, we separate the size of the hidden layers from the size of"
|
|
" vocabulary embedding.",
|
|
]
|
|
|
|
encoding = tokenizer(sequences, padding=True)
|
|
decoded_sequences = [tokenizer.decode(seq, skip_special_tokens=True) for seq in encoding["input_ids"]]
|
|
|
|
# fmt: off
|
|
expected_encoding = {
|
|
'input_ids': [
|
|
[1, 2118, 11126, 565, 35, 83, 25191, 163, 18854, 13, 12156, 12, 16101, 25376, 13807, 9, 22205, 27893, 1635, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
|
|
[1, 2118, 11126, 565, 24536, 80, 43797, 4878, 7373, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
|
|
[1, 133, 78, 65, 16, 10, 3724, 1538, 33183, 11303, 43797, 1938, 4, 870, 24165, 29105, 5, 739, 32644, 33183, 11303, 36173, 88, 80, 650, 7821, 45940, 6, 52, 2559, 5, 1836, 9, 5, 7397, 13171, 31, 5, 1836, 9, 32644, 33183, 11303, 4, 2]
|
|
],
|
|
'token_type_ids': [
|
|
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
|
|
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
|
|
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
|
|
],
|
|
'attention_mask': [
|
|
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
|
|
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
|
|
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
|
|
]
|
|
}
|
|
# fmt: on
|
|
|
|
expected_decoded_sequence = [
|
|
"ALBERT: A Lite BERT for Self-supervised Learning of Language Representations",
|
|
"ALBERT incorporates two parameter reduction techniques",
|
|
"The first one is a factorized embedding parameterization. By decomposing the large vocabulary"
|
|
" embedding matrix into two small matrices, we separate the size of the hidden layers from the size of"
|
|
" vocabulary embedding.",
|
|
]
|
|
|
|
self.assertDictEqual(encoding.data, expected_encoding)
|
|
|
|
for expected, decoded in zip(expected_decoded_sequence, decoded_sequences):
|
|
self.assertEqual(expected, decoded)
|