Spaces:
Paused
Paused
File size: 11,697 Bytes
ee6e328 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 |
# coding=utf-8
# Copyright 2019 HuggingFace 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 concurrent.futures
import json
import os
import shutil
import tempfile
import unittest
from transformers import AutoTokenizer, PreTrainedTokenizerFast
from transformers.testing_utils import require_tokenizers
from ..test_tokenization_common import TokenizerTesterMixin
@require_tokenizers
class PreTrainedTokenizationFastTest(TokenizerTesterMixin, unittest.TestCase):
rust_tokenizer_class = PreTrainedTokenizerFast
test_slow_tokenizer = False
test_rust_tokenizer = True
from_pretrained_vocab_key = "tokenizer_file"
def setUp(self):
self.test_rust_tokenizer = False # because we don't have pretrained_vocab_files_map
super().setUp()
self.test_rust_tokenizer = True
model_paths = ["robot-test/dummy-tokenizer-fast", "robot-test/dummy-tokenizer-wordlevel"]
self.bytelevel_bpe_model_name = "SaulLu/dummy-tokenizer-bytelevel-bpe"
# Inclusion of 2 tokenizers to test different types of models (Unigram and WordLevel for the moment)
self.tokenizers_list = [(PreTrainedTokenizerFast, model_path, {}) for model_path in model_paths]
tokenizer = PreTrainedTokenizerFast.from_pretrained(model_paths[0])
tokenizer.save_pretrained(self.tmpdirname)
def test_tokenizer_mismatch_warning(self):
# We disable this test for PreTrainedTokenizerFast because it is the only tokenizer that is not linked to any
# model
pass
def test_pretrained_model_lists(self):
# We disable this test for PreTrainedTokenizerFast because it is the only tokenizer that is not linked to any
# model
pass
def test_prepare_for_model(self):
# We disable this test for PreTrainedTokenizerFast because it is the only tokenizer that is not linked to any
# model
pass
def test_rust_tokenizer_signature(self):
# PreTrainedTokenizerFast doesn't have tokenizer_file in its signature
pass
def test_training_new_tokenizer(self):
tmpdirname_orig = self.tmpdirname
# Here we want to test the 2 available tokenizers that use 2 different types of models: Unigram and WordLevel.
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(f"{tokenizer.__class__.__name__} ({pretrained_name})"):
try:
self.tmpdirname = tempfile.mkdtemp()
tokenizer = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs)
tokenizer.save_pretrained(self.tmpdirname)
super().test_training_new_tokenizer()
finally:
# Even if the test fails, we must be sure that the folder is deleted and that the default tokenizer
# is restored
shutil.rmtree(self.tmpdirname)
self.tmpdirname = tmpdirname_orig
def test_training_new_tokenizer_with_special_tokens_change(self):
tmpdirname_orig = self.tmpdirname
# Here we want to test the 2 available tokenizers that use 2 different types of models: Unigram and WordLevel.
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(f"{tokenizer.__class__.__name__} ({pretrained_name})"):
try:
self.tmpdirname = tempfile.mkdtemp()
tokenizer = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs)
tokenizer.save_pretrained(self.tmpdirname)
super().test_training_new_tokenizer_with_special_tokens_change()
finally:
# Even if the test fails, we must be sure that the folder is deleted and that the default tokenizer
# is restored
shutil.rmtree(self.tmpdirname)
self.tmpdirname = tmpdirname_orig
def test_training_new_tokenizer_with_bytelevel(self):
tokenizer = self.rust_tokenizer_class.from_pretrained(self.bytelevel_bpe_model_name)
toy_text_iterator = ("a" for _ in range(1000))
new_tokenizer = tokenizer.train_new_from_iterator(text_iterator=toy_text_iterator, length=1000, vocab_size=50)
encoding_ids = new_tokenizer.encode("a🤗")
self.assertEqual(encoding_ids, [64, 172, 253, 97, 245])
def test_init_from_tokenizers_model(self):
from tokenizers import Tokenizer
sentences = ["Hello, y'all!", "How are you 😁 ? There should not be any issue right?"]
tokenizer = Tokenizer.from_pretrained("t5-base")
# Enable padding
tokenizer.enable_padding(pad_id=0, pad_token="<pad>", length=512, pad_to_multiple_of=8)
self.assertEqual(
tokenizer.padding,
{
"length": 512,
"pad_to_multiple_of": 8,
"pad_id": 0,
"pad_token": "<pad>",
"pad_type_id": 0,
"direction": "right",
},
)
fast_tokenizer = PreTrainedTokenizerFast(tokenizer_object=tokenizer)
tmpdirname = tempfile.mkdtemp()
fast_tokenizer.save_pretrained(tmpdirname)
fast_from_saved = PreTrainedTokenizerFast.from_pretrained(tmpdirname)
for tok in [fast_tokenizer, fast_from_saved]:
self.assertEqual(tok.pad_token_id, 0)
self.assertEqual(tok.padding_side, "right")
self.assertEqual(tok.pad_token, "<pad>")
self.assertEqual(tok.init_kwargs["max_length"], 512)
self.assertEqual(tok.init_kwargs["pad_to_multiple_of"], 8)
# fmt: off
self.assertEqual(tok(sentences, padding = True), {'input_ids': [[8774, 6, 3, 63, 31, 1748, 55, 1, 0, 0, 0, 0,0, 0, 0, 0],[ 571, 33, 25, 3, 2, 3, 58, 290, 225, 59, 36, 136, 962, 269, 58, 1]], '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]], 'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]})
# fmt: on
tokenizer.enable_truncation(8, stride=0, strategy="longest_first", direction="right")
self.assertEqual(
tokenizer.truncation, {"max_length": 8, "stride": 0, "strategy": "longest_first", "direction": "right"}
)
fast_tokenizer = PreTrainedTokenizerFast(tokenizer_object=tokenizer)
tmpdirname = tempfile.mkdtemp()
fast_tokenizer.save_pretrained(tmpdirname)
fast_from_saved = PreTrainedTokenizerFast.from_pretrained(tmpdirname)
for tok in [fast_tokenizer, fast_from_saved]:
self.assertEqual(tok.truncation_side, "right")
self.assertEqual(tok.init_kwargs["truncation_strategy"], "longest_first")
self.assertEqual(tok.init_kwargs["max_length"], 8)
self.assertEqual(tok.init_kwargs["stride"], 0)
# NOTE even if the model has a default max_length, it is not used...
# thus tok(sentences, truncation = True) does nothing and does not warn either
# fmt: off
self.assertEqual(tok(sentences, truncation = True, max_length = 8), {'input_ids': [[8774, 6, 3, 63, 31, 1748, 55, 1],[ 571, 33, 25, 3, 2, 3, 58, 1]], 'token_type_ids': [[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]]})
# fmt: on
@require_tokenizers
class TokenizerVersioningTest(unittest.TestCase):
def test_local_versioning(self):
tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
json_tokenizer = json.loads(tokenizer._tokenizer.to_str())
json_tokenizer["model"]["vocab"]["huggingface"] = len(tokenizer)
with tempfile.TemporaryDirectory() as tmp_dir:
# Hack to save this in the tokenizer_config.json
tokenizer.init_kwargs["fast_tokenizer_files"] = ["tokenizer.4.0.0.json"]
tokenizer.save_pretrained(tmp_dir)
json.dump(json_tokenizer, open(os.path.join(tmp_dir, "tokenizer.4.0.0.json"), "w"))
# This should pick the new tokenizer file as the version of Transformers is > 4.0.0
new_tokenizer = AutoTokenizer.from_pretrained(tmp_dir)
self.assertEqual(len(new_tokenizer), len(tokenizer) + 1)
json_tokenizer = json.loads(new_tokenizer._tokenizer.to_str())
self.assertIn("huggingface", json_tokenizer["model"]["vocab"])
# Will need to be adjusted if we reach v42 and this test is still here.
# Should pick the old tokenizer file as the version of Transformers is < 4.0.0
shutil.move(os.path.join(tmp_dir, "tokenizer.4.0.0.json"), os.path.join(tmp_dir, "tokenizer.42.0.0.json"))
tokenizer.init_kwargs["fast_tokenizer_files"] = ["tokenizer.42.0.0.json"]
tokenizer.save_pretrained(tmp_dir)
new_tokenizer = AutoTokenizer.from_pretrained(tmp_dir)
self.assertEqual(len(new_tokenizer), len(tokenizer))
json_tokenizer = json.loads(new_tokenizer._tokenizer.to_str())
self.assertNotIn("huggingface", json_tokenizer["model"]["vocab"])
def test_repo_versioning(self):
# This repo has two tokenizer files, one for v4.0.0 and above with an added token, one for versions lower.
repo = "hf-internal-testing/test-two-tokenizers"
# This should pick the new tokenizer file as the version of Transformers is > 4.0.0
tokenizer = AutoTokenizer.from_pretrained(repo)
self.assertEqual(len(tokenizer), 28997)
json_tokenizer = json.loads(tokenizer._tokenizer.to_str())
self.assertIn("huggingface", json_tokenizer["model"]["vocab"])
# Testing an older version by monkey-patching the version in the module it's used.
import transformers as old_transformers
old_transformers.tokenization_utils_base.__version__ = "3.0.0"
old_tokenizer = old_transformers.models.auto.AutoTokenizer.from_pretrained(repo)
self.assertEqual(len(old_tokenizer), 28996)
json_tokenizer = json.loads(old_tokenizer._tokenizer.to_str())
self.assertNotIn("huggingface", json_tokenizer["model"]["vocab"])
@require_tokenizers
class ReduceMutableBorrowTests(unittest.TestCase):
def test_async_share_tokenizer(self):
# See https://github.com/huggingface/transformers/pull/12550
# and https://github.com/huggingface/tokenizers/issues/537
tokenizer = PreTrainedTokenizerFast.from_pretrained("robot-test/dummy-tokenizer-wordlevel")
text = "The Matrix is a 1999 science fiction action film."
with concurrent.futures.ThreadPoolExecutor() as executor:
futures = [executor.submit(self.fetch, tokenizer, text) for i in range(10)]
return_value = [future.result() for future in futures]
self.assertEqual(return_value, [[1, 10, 0, 8, 0, 18, 0, 0, 0, 2] for i in range(10)])
def fetch(self, tokenizer, text):
return tokenizer.encode(text, truncation="longest_first", padding="longest")
|