wehub-resource-sync
303 lines
13 KiB
Python
303 lines
13 KiB
Python
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
|
|
# Copyright 2020 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 json
|
|
import os
|
|
import unittest
|
|
|
|
from paddle.utils import try_import
|
|
|
|
from paddlenlp.transformers import BlenderbotTokenizer
|
|
|
|
from ..test_tokenizer_common import TokenizerTesterMixin
|
|
|
|
VOCAB_FILES_NAMES = {
|
|
"vocab_file": "vocab.json",
|
|
"merges_file": "merges.txt",
|
|
}
|
|
|
|
|
|
class TestTokenizationBlenderbot(TokenizerTesterMixin, unittest.TestCase):
|
|
tokenizer_class = BlenderbotTokenizer
|
|
test_rust_tokenizer = False
|
|
test_offsets = False
|
|
|
|
def setUp(self):
|
|
super().setUp()
|
|
vocab = [
|
|
"l",
|
|
"o",
|
|
"w",
|
|
"e",
|
|
"r",
|
|
"s",
|
|
"t",
|
|
"i",
|
|
"d",
|
|
"n",
|
|
"\u0120",
|
|
"\u0120l",
|
|
"\u0120n",
|
|
"\u0120lo",
|
|
"\u0120low",
|
|
"er",
|
|
"\u0120lowest",
|
|
"\u0120newer",
|
|
"\u0120wider",
|
|
"<unk>",
|
|
"<s>",
|
|
"</s>",
|
|
"<pad>",
|
|
"<mask>",
|
|
]
|
|
vocab_tokens = dict(zip(vocab, range(len(vocab))))
|
|
merges = ["#version: 0.2", "\u0120low er", ""]
|
|
|
|
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"])
|
|
self.special_tokens_map = {
|
|
"bos_token": "<s>",
|
|
"eos_token": "</s>",
|
|
"cls_token": "<s>",
|
|
"sep_token": "</s>",
|
|
"unk_token": "<unk>",
|
|
"pad_token": "<pad>",
|
|
"mask_token": "<mask>",
|
|
}
|
|
|
|
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_clean_sequence(self, tokenizer, with_prefix_space=False, max_length=20, min_length=5):
|
|
toks = [(i, tokenizer.decode([i], clean_up_tokenization_spaces=False)) for i in range(len(tokenizer))]
|
|
# filter the english only character
|
|
re = try_import("regex")
|
|
if self.only_english_character:
|
|
toks = list(filter(lambda t: re.match(r"^[ a-zA-Z]+$", t[1]), toks))
|
|
toks = list(
|
|
filter(
|
|
lambda t: t[0]
|
|
== tokenizer.encode(t[1], return_token_type_ids=None, add_special_tokens=False)["input_ids"][
|
|
1
|
|
], # first is add_prefix
|
|
toks,
|
|
)
|
|
)
|
|
if max_length is not None and len(toks) > max_length:
|
|
toks = toks[:max_length]
|
|
if min_length is not None and len(toks) < min_length and len(toks) > 0:
|
|
while len(toks) < min_length:
|
|
toks = toks + toks
|
|
toks_ids = [t[0] for t in toks]
|
|
|
|
# Ensure consistency
|
|
output_txt = tokenizer.decode(toks_ids, clean_up_tokenization_spaces=False)
|
|
if " " not in output_txt and len(toks_ids) > 1:
|
|
output_txt = (
|
|
tokenizer.decode([toks_ids[0]], clean_up_tokenization_spaces=False)
|
|
+ " "
|
|
+ tokenizer.decode(toks_ids[1:], clean_up_tokenization_spaces=False)
|
|
)
|
|
if with_prefix_space:
|
|
output_txt = " " + output_txt
|
|
output_ids = tokenizer.encode(output_txt, return_token_type_ids=None, add_special_tokens=False)["input_ids"]
|
|
return output_txt, output_ids
|
|
|
|
def test_add_special_tokens(self):
|
|
tokenizers = self.get_tokenizers(do_lower_case=False)
|
|
for tokenizer in tokenizers:
|
|
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
|
input_text, ids = self.get_clean_sequence(tokenizer)
|
|
special_token = "[SPECIAL_TOKEN]"
|
|
|
|
tokenizer.add_special_tokens({"cls_token": special_token})
|
|
encoded_special_token = tokenizer.encode(
|
|
special_token, return_token_type_ids=None, add_special_tokens=False
|
|
)["input_ids"]
|
|
self.assertEqual(len(encoded_special_token), 1)
|
|
|
|
text = tokenizer.decode(ids + encoded_special_token, clean_up_tokenization_spaces=False)
|
|
encoded = tokenizer.encode(text, return_token_type_ids=None, add_special_tokens=False)["input_ids"]
|
|
print(text)
|
|
input_encoded = tokenizer.encode(input_text, return_token_type_ids=None, add_special_tokens=False)[
|
|
"input_ids"
|
|
]
|
|
special_token_id = tokenizer.encode(
|
|
special_token, return_token_type_ids=None, add_special_tokens=False
|
|
)["input_ids"]
|
|
# Each encoding adds a space token at the beginning of the sentence
|
|
self.assertNotEqual(encoded, input_encoded + special_token_id)
|
|
|
|
decoded = tokenizer.decode(encoded, skip_special_tokens=True)
|
|
self.assertTrue(special_token not in decoded)
|
|
|
|
def test_internal_consistency(self):
|
|
tokenizers = self.get_tokenizers()
|
|
for tokenizer in tokenizers:
|
|
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
|
input_text, output_text = self.get_input_output_texts(tokenizer)
|
|
|
|
tokens = tokenizer.tokenize(input_text)
|
|
ids = tokenizer.convert_tokens_to_ids(tokens)
|
|
ids_2 = tokenizer.encode(input_text, return_token_type_ids=None, add_special_tokens=False)["input_ids"]
|
|
self.assertListEqual(ids, ids_2)
|
|
|
|
tokens_2 = tokenizer.convert_ids_to_tokens(ids)
|
|
self.assertNotEqual(len(tokens_2), 0)
|
|
text_2 = tokenizer.decode(ids)
|
|
self.assertIsInstance(text_2, str)
|
|
# Each encoding adds a space token
|
|
self.assertEqual(text_2, " " + output_text)
|
|
|
|
def test_special_tokens_mask(self):
|
|
tokenizers = self.get_tokenizers(do_lower_case=False)
|
|
for tokenizer in tokenizers:
|
|
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
|
sequence_0 = "Encode this."
|
|
# Testing single inputs
|
|
encoded_sequence = tokenizer.encode(sequence_0, return_token_type_ids=None, add_special_tokens=False)[
|
|
"input_ids"
|
|
]
|
|
encoded_sequence_dict = tokenizer.encode(
|
|
sequence_0, add_special_tokens=True, return_special_tokens_mask=True # , add_prefix_space=False
|
|
)
|
|
# Each encoding adds a space token at the beginning of the sentence
|
|
encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
|
|
special_tokens_mask = [0] + encoded_sequence_dict["special_tokens_mask"]
|
|
self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
|
|
filtered_sequence = [x for i, x in enumerate(encoded_sequence_w_special) if not special_tokens_mask[i]]
|
|
# Each encoding adds a eos token at the end of the sentence
|
|
self.assertEqual(encoded_sequence, filtered_sequence[:-1])
|
|
|
|
def test_padding_to_multiple_of(self):
|
|
# token_type_ids is shorter than input_ids
|
|
tokenizers = self.get_tokenizers()
|
|
for tokenizer in tokenizers:
|
|
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
|
if tokenizer.pad_token is None:
|
|
self.skipTest("No padding token.")
|
|
else:
|
|
empty_tokens = tokenizer("", padding=True, pad_to_multiple_of=8)
|
|
normal_tokens = tokenizer("This is a sample input", padding=True, pad_to_multiple_of=8)
|
|
self.assertEqual(len(empty_tokens["input_ids"]) % 8, 0, "BatchEncoding is not multiple of 8")
|
|
self.assertEqual(len(normal_tokens["input_ids"]) % 8, 0, "BatchEncoding is not multiple of 8")
|
|
|
|
normal_tokens = tokenizer("This", pad_to_multiple_of=8)
|
|
for key, value in normal_tokens.items():
|
|
self.assertNotEqual(len(value) % 8, 0, "BatchEncoding is not multiple of 8")
|
|
|
|
# Should also work with truncation
|
|
normal_tokens = tokenizer("This", padding=True, truncation=True, pad_to_multiple_of=8)
|
|
self.assertEqual(len(normal_tokens["input_ids"]) % 8, 0, "BatchEncoding is not multiple of 8")
|
|
|
|
# truncation to something which is not a multiple of pad_to_multiple_of raises an error
|
|
self.assertRaises(
|
|
ValueError,
|
|
tokenizer.__call__,
|
|
"This",
|
|
padding=True,
|
|
truncation=True,
|
|
max_length=12,
|
|
pad_to_multiple_of=8,
|
|
)
|
|
|
|
def test_consecutive_unk_string(self):
|
|
tokenizers = self.get_tokenizers(fast=True, do_lower_case=True)
|
|
for tokenizer in tokenizers:
|
|
tokens = [tokenizer.unk_token for _ in range(2)]
|
|
string = tokenizer.convert_tokens_to_string(tokens)
|
|
encoding = tokenizer(
|
|
text=string,
|
|
runcation=True,
|
|
return_offsets_mapping=True,
|
|
)
|
|
self.assertEqual(len(encoding["input_ids"]), 4)
|
|
# offset_mapping is shorter than input_ids
|
|
self.assertEqual(len(encoding["offset_mapping"]), 4 - 1)
|
|
|
|
def test_add_tokens_tokenizer(self):
|
|
tokenizers = self.get_tokenizers(do_lower_case=False)
|
|
for tokenizer in tokenizers:
|
|
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
|
vocab_size = tokenizer.vocab_size
|
|
all_size = len(tokenizer)
|
|
|
|
self.assertNotEqual(vocab_size, 0)
|
|
|
|
# We usually have added tokens from the start in tests because our vocab fixtures are
|
|
# smaller than the original vocabs - let's not assert this
|
|
# self.assertEqual(vocab_size, all_size)
|
|
|
|
new_toks = ["aaaaa bbbbbb", "cccccccccdddddddd"]
|
|
added_toks = tokenizer.add_tokens(new_toks)
|
|
vocab_size_2 = tokenizer.vocab_size
|
|
all_size_2 = len(tokenizer)
|
|
|
|
self.assertNotEqual(vocab_size_2, 0)
|
|
self.assertEqual(vocab_size, vocab_size_2)
|
|
self.assertEqual(added_toks, len(new_toks))
|
|
self.assertEqual(all_size_2, all_size + len(new_toks))
|
|
|
|
tokens = tokenizer.encode(
|
|
"aaaaa bbbbbb low cccccccccdddddddd l", return_token_type_ids=None, add_special_tokens=False
|
|
)["input_ids"]
|
|
self.assertGreaterEqual(len(tokens), 4)
|
|
self.assertGreater(tokens[0], tokenizer.vocab_size - 1)
|
|
self.assertGreater(tokens[-2], tokenizer.vocab_size - 1)
|
|
|
|
new_toks_2 = {"eos_token": ">>>>|||<||<<|<<", "pad_token": "<<<<<|||>|>>>>|>"}
|
|
added_toks_2 = tokenizer.add_special_tokens(new_toks_2)
|
|
vocab_size_3 = tokenizer.vocab_size
|
|
all_size_3 = len(tokenizer)
|
|
|
|
self.assertNotEqual(vocab_size_3, 0)
|
|
self.assertEqual(vocab_size, vocab_size_3)
|
|
self.assertEqual(added_toks_2, len(new_toks_2))
|
|
self.assertEqual(all_size_3, all_size_2 + len(new_toks_2))
|
|
|
|
tokens = tokenizer.encode(
|
|
">>>>|||<||<<|<< aaaaa bbbbbb low cccccccccdddddddd <<<<<|||>|>>>>|> l",
|
|
return_token_type_ids=None,
|
|
add_special_tokens=False,
|
|
)["input_ids"]
|
|
self.assertGreaterEqual(len(tokens), 6)
|
|
self.assertGreater(tokens[0], tokenizer.vocab_size - 1)
|
|
self.assertGreater(tokens[0], tokens[1])
|
|
self.assertGreater(tokens[-2], tokenizer.vocab_size - 1)
|
|
self.assertGreater(tokens[-2], tokens[-3])
|
|
self.assertEqual(tokens[0], tokenizer.eos_token_id)
|
|
self.assertEqual(tokens[-2], tokenizer.pad_token_id)
|
|
|
|
def test_special_tokens_mask_input_pairs(self):
|
|
# token_ids_1 Will be ignored
|
|
self.skipTest("token_ids_1 Will be ignored")
|
|
|
|
def test_number_of_added_tokens(self):
|
|
# token_ids_1 Will be ignored
|
|
self.skipTest("token_ids_1 Will be ignored")
|
|
|
|
def test_mask_output(self):
|
|
# token_ids_1 Will be ignored
|
|
self.skipTest("token_ids_1 Will be ignored")
|
|
|
|
def test_maximum_encoding_length_pair_input(self):
|
|
# token_ids_1 Will be ignored
|
|
self.skipTest("token_ids_1 Will be ignored")
|