[Test refactor 1/5] Per-folder tests reorganization (#15725)
* Per-folder tests reorganization Co-authored-by: sgugger <sylvain.gugger@gmail.com> Co-authored-by: Stas Bekman <stas@stason.org>
This commit is contained in:
Lysandre Debut
0
tests/tokenization/__init__.py
Normal file
0
tests/tokenization/__init__.py
Normal file
166
tests/tokenization/test_tokenization_fast.py
Normal file
166
tests/tokenization/test_tokenization_fast.py
Normal file
@@ -0,0 +1,166 @@
|
||||
# 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"]
|
||||
|
||||
# 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
|
||||
|
||||
|
||||
@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")
|
||||
287
tests/tokenization/test_tokenization_utils.py
Normal file
287
tests/tokenization/test_tokenization_utils.py
Normal file
@@ -0,0 +1,287 @@
|
||||
# coding=utf-8
|
||||
# Copyright 2018 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.
|
||||
"""
|
||||
isort:skip_file
|
||||
"""
|
||||
import os
|
||||
import pickle
|
||||
import tempfile
|
||||
import unittest
|
||||
from typing import Callable, Optional
|
||||
|
||||
import numpy as np
|
||||
|
||||
# Ensure there are no circular imports when importing the parent class
|
||||
from transformers import PreTrainedTokenizerFast
|
||||
|
||||
from transformers import (
|
||||
BatchEncoding,
|
||||
BertTokenizer,
|
||||
BertTokenizerFast,
|
||||
PreTrainedTokenizer,
|
||||
TensorType,
|
||||
TokenSpan,
|
||||
is_tokenizers_available,
|
||||
)
|
||||
from transformers.models.gpt2.tokenization_gpt2 import GPT2Tokenizer
|
||||
from transformers.testing_utils import CaptureStderr, require_flax, require_tf, require_tokenizers, require_torch, slow
|
||||
|
||||
|
||||
if is_tokenizers_available():
|
||||
from tokenizers import Tokenizer
|
||||
from tokenizers.models import WordPiece
|
||||
|
||||
|
||||
class TokenizerUtilsTest(unittest.TestCase):
|
||||
def check_tokenizer_from_pretrained(self, tokenizer_class):
|
||||
s3_models = list(tokenizer_class.max_model_input_sizes.keys())
|
||||
for model_name in s3_models[:1]:
|
||||
tokenizer = tokenizer_class.from_pretrained(model_name)
|
||||
self.assertIsNotNone(tokenizer)
|
||||
self.assertIsInstance(tokenizer, tokenizer_class)
|
||||
self.assertIsInstance(tokenizer, PreTrainedTokenizer)
|
||||
|
||||
for special_tok in tokenizer.all_special_tokens:
|
||||
self.assertIsInstance(special_tok, str)
|
||||
special_tok_id = tokenizer.convert_tokens_to_ids(special_tok)
|
||||
self.assertIsInstance(special_tok_id, int)
|
||||
|
||||
def assert_dump_and_restore(self, be_original: BatchEncoding, equal_op: Optional[Callable] = None):
|
||||
batch_encoding_str = pickle.dumps(be_original)
|
||||
self.assertIsNotNone(batch_encoding_str)
|
||||
|
||||
be_restored = pickle.loads(batch_encoding_str)
|
||||
|
||||
# Ensure is_fast is correctly restored
|
||||
self.assertEqual(be_restored.is_fast, be_original.is_fast)
|
||||
|
||||
# Ensure encodings are potentially correctly restored
|
||||
if be_original.is_fast:
|
||||
self.assertIsNotNone(be_restored.encodings)
|
||||
else:
|
||||
self.assertIsNone(be_restored.encodings)
|
||||
|
||||
# Ensure the keys are the same
|
||||
for original_v, restored_v in zip(be_original.values(), be_restored.values()):
|
||||
if equal_op:
|
||||
self.assertTrue(equal_op(restored_v, original_v))
|
||||
else:
|
||||
self.assertEqual(restored_v, original_v)
|
||||
|
||||
@slow
|
||||
def test_pretrained_tokenizers(self):
|
||||
self.check_tokenizer_from_pretrained(GPT2Tokenizer)
|
||||
|
||||
def test_tensor_type_from_str(self):
|
||||
self.assertEqual(TensorType("tf"), TensorType.TENSORFLOW)
|
||||
self.assertEqual(TensorType("pt"), TensorType.PYTORCH)
|
||||
self.assertEqual(TensorType("np"), TensorType.NUMPY)
|
||||
|
||||
@require_tokenizers
|
||||
def test_batch_encoding_pickle(self):
|
||||
import numpy as np
|
||||
|
||||
tokenizer_p = BertTokenizer.from_pretrained("bert-base-cased")
|
||||
tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")
|
||||
|
||||
# Python no tensor
|
||||
with self.subTest("BatchEncoding (Python, return_tensors=None)"):
|
||||
self.assert_dump_and_restore(tokenizer_p("Small example to encode"))
|
||||
|
||||
with self.subTest("BatchEncoding (Python, return_tensors=NUMPY)"):
|
||||
self.assert_dump_and_restore(
|
||||
tokenizer_p("Small example to encode", return_tensors=TensorType.NUMPY), np.array_equal
|
||||
)
|
||||
|
||||
with self.subTest("BatchEncoding (Rust, return_tensors=None)"):
|
||||
self.assert_dump_and_restore(tokenizer_r("Small example to encode"))
|
||||
|
||||
with self.subTest("BatchEncoding (Rust, return_tensors=NUMPY)"):
|
||||
self.assert_dump_and_restore(
|
||||
tokenizer_r("Small example to encode", return_tensors=TensorType.NUMPY), np.array_equal
|
||||
)
|
||||
|
||||
@require_tf
|
||||
@require_tokenizers
|
||||
def test_batch_encoding_pickle_tf(self):
|
||||
import tensorflow as tf
|
||||
|
||||
def tf_array_equals(t1, t2):
|
||||
return tf.reduce_all(tf.equal(t1, t2))
|
||||
|
||||
tokenizer_p = BertTokenizer.from_pretrained("bert-base-cased")
|
||||
tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")
|
||||
|
||||
with self.subTest("BatchEncoding (Python, return_tensors=TENSORFLOW)"):
|
||||
self.assert_dump_and_restore(
|
||||
tokenizer_p("Small example to encode", return_tensors=TensorType.TENSORFLOW), tf_array_equals
|
||||
)
|
||||
|
||||
with self.subTest("BatchEncoding (Rust, return_tensors=TENSORFLOW)"):
|
||||
self.assert_dump_and_restore(
|
||||
tokenizer_r("Small example to encode", return_tensors=TensorType.TENSORFLOW), tf_array_equals
|
||||
)
|
||||
|
||||
@require_torch
|
||||
@require_tokenizers
|
||||
def test_batch_encoding_pickle_pt(self):
|
||||
import torch
|
||||
|
||||
tokenizer_p = BertTokenizer.from_pretrained("bert-base-cased")
|
||||
tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")
|
||||
|
||||
with self.subTest("BatchEncoding (Python, return_tensors=PYTORCH)"):
|
||||
self.assert_dump_and_restore(
|
||||
tokenizer_p("Small example to encode", return_tensors=TensorType.PYTORCH), torch.equal
|
||||
)
|
||||
|
||||
with self.subTest("BatchEncoding (Rust, return_tensors=PYTORCH)"):
|
||||
self.assert_dump_and_restore(
|
||||
tokenizer_r("Small example to encode", return_tensors=TensorType.PYTORCH), torch.equal
|
||||
)
|
||||
|
||||
@require_tokenizers
|
||||
def test_batch_encoding_is_fast(self):
|
||||
tokenizer_p = BertTokenizer.from_pretrained("bert-base-cased")
|
||||
tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")
|
||||
|
||||
with self.subTest("Python Tokenizer"):
|
||||
self.assertFalse(tokenizer_p("Small example to_encode").is_fast)
|
||||
|
||||
with self.subTest("Rust Tokenizer"):
|
||||
self.assertTrue(tokenizer_r("Small example to_encode").is_fast)
|
||||
|
||||
@require_tokenizers
|
||||
def test_batch_encoding_word_to_tokens(self):
|
||||
tokenizer_r = BertTokenizerFast.from_pretrained("bert-base-cased")
|
||||
encoded = tokenizer_r(["Test", "\xad", "test"], is_split_into_words=True)
|
||||
|
||||
self.assertEqual(encoded.word_to_tokens(0), TokenSpan(start=1, end=2))
|
||||
self.assertEqual(encoded.word_to_tokens(1), None)
|
||||
self.assertEqual(encoded.word_to_tokens(2), TokenSpan(start=2, end=3))
|
||||
|
||||
def test_batch_encoding_with_labels(self):
|
||||
batch = BatchEncoding({"inputs": [[1, 2, 3], [4, 5, 6]], "labels": [0, 1]})
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="np")
|
||||
self.assertEqual(tensor_batch["inputs"].shape, (2, 3))
|
||||
self.assertEqual(tensor_batch["labels"].shape, (2,))
|
||||
# test converting the converted
|
||||
with CaptureStderr() as cs:
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="np")
|
||||
self.assertFalse(len(cs.err), msg=f"should have no warning, but got {cs.err}")
|
||||
|
||||
batch = BatchEncoding({"inputs": [1, 2, 3], "labels": 0})
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="np", prepend_batch_axis=True)
|
||||
self.assertEqual(tensor_batch["inputs"].shape, (1, 3))
|
||||
self.assertEqual(tensor_batch["labels"].shape, (1,))
|
||||
|
||||
@require_torch
|
||||
def test_batch_encoding_with_labels_pt(self):
|
||||
batch = BatchEncoding({"inputs": [[1, 2, 3], [4, 5, 6]], "labels": [0, 1]})
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="pt")
|
||||
self.assertEqual(tensor_batch["inputs"].shape, (2, 3))
|
||||
self.assertEqual(tensor_batch["labels"].shape, (2,))
|
||||
# test converting the converted
|
||||
with CaptureStderr() as cs:
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="pt")
|
||||
self.assertFalse(len(cs.err), msg=f"should have no warning, but got {cs.err}")
|
||||
|
||||
batch = BatchEncoding({"inputs": [1, 2, 3], "labels": 0})
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="pt", prepend_batch_axis=True)
|
||||
self.assertEqual(tensor_batch["inputs"].shape, (1, 3))
|
||||
self.assertEqual(tensor_batch["labels"].shape, (1,))
|
||||
|
||||
@require_tf
|
||||
def test_batch_encoding_with_labels_tf(self):
|
||||
batch = BatchEncoding({"inputs": [[1, 2, 3], [4, 5, 6]], "labels": [0, 1]})
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="tf")
|
||||
self.assertEqual(tensor_batch["inputs"].shape, (2, 3))
|
||||
self.assertEqual(tensor_batch["labels"].shape, (2,))
|
||||
# test converting the converted
|
||||
with CaptureStderr() as cs:
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="tf")
|
||||
self.assertFalse(len(cs.err), msg=f"should have no warning, but got {cs.err}")
|
||||
|
||||
batch = BatchEncoding({"inputs": [1, 2, 3], "labels": 0})
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="tf", prepend_batch_axis=True)
|
||||
self.assertEqual(tensor_batch["inputs"].shape, (1, 3))
|
||||
self.assertEqual(tensor_batch["labels"].shape, (1,))
|
||||
|
||||
@require_flax
|
||||
def test_batch_encoding_with_labels_jax(self):
|
||||
batch = BatchEncoding({"inputs": [[1, 2, 3], [4, 5, 6]], "labels": [0, 1]})
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="jax")
|
||||
self.assertEqual(tensor_batch["inputs"].shape, (2, 3))
|
||||
self.assertEqual(tensor_batch["labels"].shape, (2,))
|
||||
# test converting the converted
|
||||
with CaptureStderr() as cs:
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="jax")
|
||||
self.assertFalse(len(cs.err), msg=f"should have no warning, but got {cs.err}")
|
||||
|
||||
batch = BatchEncoding({"inputs": [1, 2, 3], "labels": 0})
|
||||
tensor_batch = batch.convert_to_tensors(tensor_type="jax", prepend_batch_axis=True)
|
||||
self.assertEqual(tensor_batch["inputs"].shape, (1, 3))
|
||||
self.assertEqual(tensor_batch["labels"].shape, (1,))
|
||||
|
||||
def test_padding_accepts_tensors(self):
|
||||
features = [{"input_ids": np.array([0, 1, 2])}, {"input_ids": np.array([0, 1, 2, 3])}]
|
||||
tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
|
||||
|
||||
batch = tokenizer.pad(features, padding=True)
|
||||
self.assertTrue(isinstance(batch["input_ids"], np.ndarray))
|
||||
self.assertEqual(batch["input_ids"].tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
|
||||
batch = tokenizer.pad(features, padding=True, return_tensors="np")
|
||||
self.assertTrue(isinstance(batch["input_ids"], np.ndarray))
|
||||
self.assertEqual(batch["input_ids"].tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
|
||||
|
||||
@require_torch
|
||||
def test_padding_accepts_tensors_pt(self):
|
||||
import torch
|
||||
|
||||
features = [{"input_ids": torch.tensor([0, 1, 2])}, {"input_ids": torch.tensor([0, 1, 2, 3])}]
|
||||
tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
|
||||
|
||||
batch = tokenizer.pad(features, padding=True)
|
||||
self.assertTrue(isinstance(batch["input_ids"], torch.Tensor))
|
||||
self.assertEqual(batch["input_ids"].tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
|
||||
batch = tokenizer.pad(features, padding=True, return_tensors="pt")
|
||||
self.assertTrue(isinstance(batch["input_ids"], torch.Tensor))
|
||||
self.assertEqual(batch["input_ids"].tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
|
||||
|
||||
@require_tf
|
||||
def test_padding_accepts_tensors_tf(self):
|
||||
import tensorflow as tf
|
||||
|
||||
features = [{"input_ids": tf.constant([0, 1, 2])}, {"input_ids": tf.constant([0, 1, 2, 3])}]
|
||||
tokenizer = BertTokenizer.from_pretrained("bert-base-cased")
|
||||
|
||||
batch = tokenizer.pad(features, padding=True)
|
||||
self.assertTrue(isinstance(batch["input_ids"], tf.Tensor))
|
||||
self.assertEqual(batch["input_ids"].numpy().tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
|
||||
batch = tokenizer.pad(features, padding=True, return_tensors="tf")
|
||||
self.assertTrue(isinstance(batch["input_ids"], tf.Tensor))
|
||||
self.assertEqual(batch["input_ids"].numpy().tolist(), [[0, 1, 2, tokenizer.pad_token_id], [0, 1, 2, 3]])
|
||||
|
||||
@require_tokenizers
|
||||
def test_instantiation_from_tokenizers(self):
|
||||
bert_tokenizer = Tokenizer(WordPiece(unk_token="[UNK]"))
|
||||
PreTrainedTokenizerFast(tokenizer_object=bert_tokenizer)
|
||||
|
||||
@require_tokenizers
|
||||
def test_instantiation_from_tokenizers_json_file(self):
|
||||
bert_tokenizer = Tokenizer(WordPiece(unk_token="[UNK]"))
|
||||
with tempfile.TemporaryDirectory() as tmpdirname:
|
||||
bert_tokenizer.save(os.path.join(tmpdirname, "tokenizer.json"))
|
||||
PreTrainedTokenizerFast(tokenizer_file=os.path.join(tmpdirname, "tokenizer.json"))
|
||||
Reference in New Issue
Block a user