EngineX-Cambricon/enginex-mlu370-any2any
9
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init

Browse Source
This commit is contained in:
luopingyi
2025-10-09 16:47:16 +08:00
parent c8feb4deb5
commit e27e3f16bb
5248 changed files with 1778505 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

0
transformers/tests/models/esm/__init__.py Normal file
View File

395
transformers/tests/models/esm/test_modeling_esm.py Normal file
View File

@@ -0,0 +1,395 @@
# Copyright 2022 The HuggingFace Inc. 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.
"""Testing suite for the PyTorch ESM model."""
import tempfile
import unittest
import pytest
from transformers import EsmConfig, is_torch_available
from transformers.testing_utils import (
TestCasePlus,
is_flaky,
require_bitsandbytes,
require_flash_attn,
require_torch,
require_torch_gpu,
slow,
torch_device,
)
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import EsmForMaskedLM, EsmForSequenceClassification, EsmForTokenClassification, EsmModel
from transformers.models.esm.modeling_esm import (
EsmEmbeddings,
create_position_ids_from_input_ids,
)
# copied from tests.test_modeling_roberta
class EsmModelTester:
def __init__(
self,
parent,
batch_size=13,
seq_length=7,
is_training=False,
use_input_mask=True,
use_token_type_ids=False,
use_labels=True,
vocab_size=33,
hidden_size=32,
num_hidden_layers=2,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
type_sequence_label_size=2,
initializer_range=0.02,
num_labels=3,
num_choices=4,
scope=None,
position_embedding_type="rotary",
):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_input_mask = use_input_mask
self.use_token_type_ids = use_token_type_ids
self.use_labels = use_labels
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
self.num_labels = num_labels
self.num_choices = num_choices
self.scope = scope
self.position_embedding_type = position_embedding_type
def prepare_config_and_inputs(self):
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = random_attention_mask([self.batch_size, self.seq_length])
sequence_labels = None
token_labels = None
choice_labels = None
if self.use_labels:
sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = self.get_config()
return config, input_ids, input_mask, sequence_labels, token_labels, choice_labels
def get_config(self):
return EsmConfig(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
pad_token_id=1,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range,
position_embedding_type=self.position_embedding_type,
)
def create_and_check_model(self, config, input_ids, input_mask, sequence_labels, token_labels, choice_labels):
model = EsmModel(config=config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=input_mask)
result = model(input_ids)
result = model(input_ids)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
self.parent.assertEqual(result.pooler_output.shape, (self.batch_size, self.hidden_size))
def create_and_check_for_masked_lm(
self, config, input_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = EsmForMaskedLM(config=config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=input_mask, labels=token_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
def create_and_check_for_token_classification(
self, config, input_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.num_labels = self.num_labels
model = EsmForTokenClassification(config=config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=input_mask, labels=token_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
def create_and_check_forward_and_backwards(
self,
config,
input_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
gradient_checkpointing=False,
):
model = EsmForMaskedLM(config)
if gradient_checkpointing:
model.gradient_checkpointing_enable()
model.to(torch_device)
result = model(input_ids, attention_mask=input_mask, labels=token_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
result.loss.backward()
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
config,
input_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
) = config_and_inputs
inputs_dict = {"input_ids": input_ids, "attention_mask": input_mask}
return config, inputs_dict
@require_torch
class EsmModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
test_mismatched_shapes = False
all_model_classes = (
(
EsmForMaskedLM,
EsmModel,
EsmForSequenceClassification,
EsmForTokenClassification,
)
if is_torch_available()
else ()
)
pipeline_model_mapping = (
{
"feature-extraction": EsmModel,
"fill-mask": EsmForMaskedLM,
"text-classification": EsmForSequenceClassification,
"token-classification": EsmForTokenClassification,
"zero-shot": EsmForSequenceClassification,
}
if is_torch_available()
else {}
)
test_sequence_classification_problem_types = True
model_split_percents = [0.5, 0.8, 0.9]
def setUp(self):
self.model_tester = EsmModelTester(self)
self.config_tester = ConfigTester(self, config_class=EsmConfig, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
def test_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_model_various_embeddings(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
config_and_inputs[0].position_embedding_type = type
config_and_inputs[0]._attn_implementation = "eager"
self.model_tester.create_and_check_model(*config_and_inputs)
def test_for_masked_lm(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_lm(*config_and_inputs)
def test_for_token_classification(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_token_classification(*config_and_inputs)
def test_esm_gradient_checkpointing(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_forward_and_backwards(*config_and_inputs, gradient_checkpointing=True)
@slow
def test_model_from_pretrained(self):
model_name = "facebook/esm2_t6_8M_UR50D"
model = EsmModel.from_pretrained(model_name)
self.assertIsNotNone(model)
def test_create_position_ids_respects_padding_index(self):
"""This is a regression test for https://github.com/huggingface/transformers/issues/1761
The position ids should be masked with the embedding object's padding index. Therefore, the
first available non-padding position index is EsmEmbeddings.padding_idx + 1
"""
config = self.model_tester.prepare_config_and_inputs()[0]
model = EsmEmbeddings(config=config)
input_ids = torch.as_tensor([[12, 31, 13, model.padding_idx]])
expected_positions = torch.as_tensor(
[
[
0 + model.padding_idx + 1,
1 + model.padding_idx + 1,
2 + model.padding_idx + 1,
model.padding_idx,
]
]
)
position_ids = create_position_ids_from_input_ids(input_ids, model.padding_idx)
self.assertEqual(position_ids.shape, expected_positions.shape)
self.assertTrue(torch.all(torch.eq(position_ids, expected_positions)))
def test_create_position_ids_from_inputs_embeds(self):
"""This is a regression test for https://github.com/huggingface/transformers/issues/1761
The position ids should be masked with the embedding object's padding index. Therefore, the
first available non-padding position index is EsmEmbeddings.padding_idx + 1
"""
config = self.model_tester.prepare_config_and_inputs()[0]
embeddings = EsmEmbeddings(config=config)
inputs_embeds = torch.empty(2, 4, 30)
expected_single_positions = [
0 + embeddings.padding_idx + 1,
1 + embeddings.padding_idx + 1,
2 + embeddings.padding_idx + 1,
3 + embeddings.padding_idx + 1,
]
expected_positions = torch.as_tensor([expected_single_positions, expected_single_positions])
position_ids = embeddings.create_position_ids_from_inputs_embeds(inputs_embeds)
self.assertEqual(position_ids.shape, expected_positions.shape)
self.assertTrue(torch.all(torch.eq(position_ids, expected_positions)))
@unittest.skip(reason="Esm does not support embedding resizing")
def test_resize_embeddings_untied(self):
pass
@unittest.skip(reason="Esm does not support embedding resizing")
def test_resize_tokens_embeddings(self):
pass
@require_flash_attn
@require_torch_gpu
@pytest.mark.flash_attn_test
@is_flaky()
@slow
def test_flash_attn_2_equivalence(self):
for model_class in self.all_model_classes:
if not model_class._supports_flash_attn:
self.skipTest(reason="Model does not support Flash Attention 2")
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
model = model_class(config)
with tempfile.TemporaryDirectory() as tmpdirname:
model.save_pretrained(tmpdirname)
model_fa = model_class.from_pretrained(
tmpdirname, dtype=torch.float16, attn_implementation="flash_attention_2"
)
model_fa.to(torch_device)
model = model_class.from_pretrained(tmpdirname, dtype=torch.float16, attn_implementation="eager")
model.to(torch_device)
dummy_input = inputs_dict[model_class.main_input_name]
dummy_input = dummy_input.to(torch_device)
outputs = model(dummy_input, output_hidden_states=True)
outputs_fa = model_fa(dummy_input, output_hidden_states=True)
logits = outputs.hidden_states[-1]
logits_fa = outputs_fa.hidden_states[-1]
torch.testing.assert_close(logits_fa, logits, atol=1e-2, rtol=1e-3)
@slow
@require_torch
class EsmModelIntegrationTest(TestCasePlus):
def test_inference_masked_lm(self):
with torch.no_grad():
model = EsmForMaskedLM.from_pretrained("facebook/esm2_t6_8M_UR50D")
model.eval()
input_ids = torch.tensor([[0, 1, 2, 3, 4, 5]])
output = model(input_ids)[0]
vocab_size = 33
expected_shape = torch.Size((1, 6, vocab_size))
self.assertEqual(output.shape, expected_shape)
expected_slice = torch.tensor(
[[[8.9215, -10.5898, -6.4671], [-6.3967, -13.9114, -1.1212], [-7.7812, -13.9516, -3.7406]]]
)
torch.testing.assert_close(output[:, :3, :3], expected_slice, rtol=1e-4, atol=1e-4)
def test_inference_no_head(self):
with torch.no_grad():
model = EsmModel.from_pretrained("facebook/esm2_t6_8M_UR50D")
model.eval()
input_ids = torch.tensor([[0, 6, 4, 13, 5, 4, 16, 12, 11, 7, 2]])
output = model(input_ids)[0]
# compare the actual values for a slice.
expected_slice = torch.tensor(
[[[0.1444, 0.5413, 0.3248], [0.3034, 0.0053, 0.3108], [0.3228, -0.2499, 0.3415]]]
)
torch.testing.assert_close(output[:, :3, :3], expected_slice, rtol=1e-4, atol=1e-4)
@require_bitsandbytes
def test_inference_bitsandbytes(self):
model = EsmForMaskedLM.from_pretrained("facebook/esm2_t36_3B_UR50D", load_in_8bit=True)
input_ids = torch.tensor([[0, 6, 4, 13, 5, 4, 16, 12, 11, 7, 2]]).to(model.device)
# Just test if inference works
with torch.no_grad():
_ = model(input_ids)[0]
model = EsmForMaskedLM.from_pretrained("facebook/esm2_t36_3B_UR50D", load_in_4bit=True)
input_ids = torch.tensor([[0, 6, 4, 13, 5, 4, 16, 12, 11, 7, 2]]).to(model.device)
# Just test if inference works
_ = model(input_ids)[0]

279
transformers/tests/models/esm/test_modeling_esmfold.py Normal file
View File

@@ -0,0 +1,279 @@
# Copyright 2022 The HuggingFace Inc. 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.
"""Testing suite for the PyTorch ESM model."""
import unittest
from transformers import EsmConfig, is_torch_available
from transformers.testing_utils import TestCasePlus, is_flaky, require_torch, slow, torch_device
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers.models.esm.modeling_esmfold import EsmForProteinFolding
class EsmFoldModelTester:
def __init__(
self,
parent,
batch_size=13,
seq_length=7,
is_training=False,
use_input_mask=True,
use_token_type_ids=False,
use_labels=False,
vocab_size=19,
hidden_size=32,
num_hidden_layers=2,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
type_sequence_label_size=2,
initializer_range=0.02,
num_labels=3,
num_choices=4,
scope=None,
):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_input_mask = use_input_mask
self.use_token_type_ids = use_token_type_ids
self.use_labels = use_labels
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
self.num_labels = num_labels
self.num_choices = num_choices
self.scope = scope
def prepare_config_and_inputs(self):
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = random_attention_mask([self.batch_size, self.seq_length])
sequence_labels = None
token_labels = None
choice_labels = None
if self.use_labels:
sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = self.get_config()
return config, input_ids, input_mask, sequence_labels, token_labels, choice_labels
def get_config(self):
esmfold_config = {
"trunk": {
"num_blocks": 2,
"sequence_state_dim": 64,
"pairwise_state_dim": 16,
"sequence_head_width": 4,
"pairwise_head_width": 4,
"position_bins": 4,
"chunk_size": 16,
"structure_module": {
"ipa_dim": 16,
"num_angles": 7,
"num_blocks": 2,
"num_heads_ipa": 4,
"pairwise_dim": 16,
"resnet_dim": 16,
"sequence_dim": 48,
},
},
"fp16_esm": False,
"lddt_head_hid_dim": 16,
}
config = EsmConfig(
vocab_size=33,
hidden_size=self.hidden_size,
pad_token_id=1,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range,
is_folding_model=True,
esmfold_config=esmfold_config,
)
return config
def create_and_check_model(self, config, input_ids, input_mask, sequence_labels, token_labels, choice_labels):
model = EsmForProteinFolding(config=config).float()
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=input_mask)
result = model(input_ids)
result = model(input_ids)
self.parent.assertEqual(result.positions.shape, (2, self.batch_size, self.seq_length, 14, 3))
self.parent.assertEqual(result.angles.shape, (2, self.batch_size, self.seq_length, 7, 2))
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
config,
input_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
) = config_and_inputs
inputs_dict = {"input_ids": input_ids, "attention_mask": input_mask}
return config, inputs_dict
@require_torch
class EsmFoldModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
test_mismatched_shapes = False
all_model_classes = (EsmForProteinFolding,) if is_torch_available() else ()
pipeline_model_mapping = {} if is_torch_available() else {}
test_sequence_classification_problem_types = False
def setUp(self):
self.model_tester = EsmFoldModelTester(self)
self.config_tester = ConfigTester(self, config_class=EsmConfig, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
def test_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
@is_flaky(
description="The computed `s = s / norm_denom` in `EsmFoldAngleResnet` is numerically instable if `norm_denom` is very small."
)
def test_batching_equivalence(self):
super().test_batching_equivalence()
@unittest.skip(reason="Does not support attention outputs")
def test_attention_outputs(self):
pass
@unittest.skip
def test_correct_missing_keys(self):
pass
@unittest.skip(reason="Esm does not support embedding resizing")
def test_resize_embeddings_untied(self):
pass
@unittest.skip(reason="Esm does not support embedding resizing")
def test_resize_tokens_embeddings(self):
pass
@unittest.skip(reason="ESMFold does not support passing input embeds!")
def test_inputs_embeds(self):
pass
@unittest.skip(reason="ESMFold does not support head pruning.")
def test_head_pruning(self):
pass
@unittest.skip(reason="ESMFold does not support head pruning.")
def test_head_pruning_integration(self):
pass
@unittest.skip(reason="ESMFold does not support head pruning.")
def test_head_pruning_save_load_from_config_init(self):
pass
@unittest.skip(reason="ESMFold does not support head pruning.")
def test_head_pruning_save_load_from_pretrained(self):
pass
@unittest.skip(reason="ESMFold does not support head pruning.")
def test_headmasking(self):
pass
@unittest.skip(reason="ESMFold does not output hidden states in the normal way.")
def test_hidden_states_output(self):
pass
@unittest.skip(reason="ESMfold does not output hidden states in the normal way.")
def test_retain_grad_hidden_states_attentions(self):
pass
@unittest.skip(reason="ESMFold only has one output format.")
def test_model_outputs_equivalence(self):
pass
@unittest.skip(reason="ESMFold does not support input chunking.")
def test_feed_forward_chunking(self):
pass
@unittest.skip(
reason="ESMFold doesn't respect you and it certainly doesn't respect your initialization arguments."
)
def test_initialization(self):
pass
@unittest.skip(reason="ESMFold doesn't support torchscript compilation.")
def test_torchscript_output_attentions(self):
pass
@unittest.skip(reason="ESMFold doesn't support torchscript compilation.")
def test_torchscript_output_hidden_state(self):
pass
@unittest.skip(reason="ESMFold doesn't support torchscript compilation.")
def test_torchscript_simple(self):
pass
@unittest.skip(reason="ESMFold doesn't support data parallel.")
def test_multi_gpu_data_parallel_forward(self):
pass
@require_torch
class EsmModelIntegrationTest(TestCasePlus):
@slow
def test_inference_protein_folding(self):
model = EsmForProteinFolding.from_pretrained("facebook/esmfold_v1").float()
model.eval()
input_ids = torch.tensor([[0, 6, 4, 13, 5, 4, 16, 12, 11, 7, 2]])
position_outputs = model(input_ids)["positions"]
expected_slice = torch.tensor([2.5828, 0.7993, -10.9334], dtype=torch.float32)
torch.testing.assert_close(position_outputs[0, 0, 0, 0], expected_slice, rtol=1e-4, atol=1e-4)

113
transformers/tests/models/esm/test_tokenization_esm.py Normal file
View File

@@ -0,0 +1,113 @@
# Copyright 2021 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 tempfile
import unittest
from transformers.models.esm.tokenization_esm import VOCAB_FILES_NAMES, EsmTokenizer
from transformers.testing_utils import require_tokenizers
from transformers.tokenization_utils import PreTrainedTokenizer
from transformers.tokenization_utils_base import PreTrainedTokenizerBase
@require_tokenizers
class ESMTokenizationTest(unittest.TestCase):
tokenizer_class = EsmTokenizer
@classmethod
def setUpClass(cls):
super().setUpClass()
cls.tmpdirname = tempfile.mkdtemp()
vocab_tokens: list[str] = ["<cls>", "<pad>", "<eos>", "<unk>", "L", "A", "G", "V", "S", "E", "R", "T", "I", "D", "P", "K", "Q", "N", "F", "Y", "M", "H", "W", "C", "X", "B", "U", "Z", "O", ".", "-", "<null_1>", "<mask>"] # fmt: skip
cls.vocab_file = os.path.join(cls.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
with open(cls.vocab_file, "w", encoding="utf-8") as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
def get_tokenizers(cls, **kwargs) -> list[PreTrainedTokenizerBase]:
return [cls.get_tokenizer(**kwargs)]
@classmethod
def get_tokenizer(cls, pretrained_name=None, **kwargs) -> PreTrainedTokenizer:
pretrained_name = pretrained_name or cls.tmpdirname
return cls.tokenizer_class.from_pretrained(pretrained_name, **kwargs)
def test_tokenizer_single_example(self):
tokenizer = self.tokenizer_class(self.vocab_file)
tokens = tokenizer.tokenize("LAGVS")
self.assertListEqual(tokens, ["L", "A", "G", "V", "S"])
self.assertListEqual(tokenizer.convert_tokens_to_ids(tokens), [4, 5, 6, 7, 8])
def test_tokenizer_encode_single(self):
tokenizer = self.tokenizer_class(self.vocab_file)
seq = "LAGVS"
self.assertListEqual(tokenizer.encode(seq), [0, 4, 5, 6, 7, 8, 2])
def test_tokenizer_call_no_pad(self):
tokenizer = self.tokenizer_class(self.vocab_file)
seq_batch = ["LAGVS", "WCB"]
tokens_batch = tokenizer(seq_batch, padding=False)["input_ids"]
self.assertListEqual(tokens_batch, [[0, 4, 5, 6, 7, 8, 2], [0, 22, 23, 25, 2]])
def test_tokenizer_call_pad(self):
tokenizer = self.tokenizer_class(self.vocab_file)
seq_batch = ["LAGVS", "WCB"]
tokens_batch = tokenizer(seq_batch, padding=True)["input_ids"]
self.assertListEqual(tokens_batch, [[0, 4, 5, 6, 7, 8, 2], [0, 22, 23, 25, 2, 1, 1]])
def test_tokenize_special_tokens(self):
"""Test `tokenize` with special tokens."""
tokenizers = self.get_tokenizers(fast=True)
for tokenizer in tokenizers:
with self.subTest(f"{tokenizer.__class__.__name__}"):
SPECIAL_TOKEN_1 = "<unk>"
SPECIAL_TOKEN_2 = "<mask>"
token_1 = tokenizer.tokenize(SPECIAL_TOKEN_1)
token_2 = tokenizer.tokenize(SPECIAL_TOKEN_2)
self.assertEqual(len(token_1), 1)
self.assertEqual(len(token_2), 1)
self.assertEqual(token_1[0], SPECIAL_TOKEN_1)
self.assertEqual(token_2[0], SPECIAL_TOKEN_2)
def test_add_tokens(self):
tokenizer = self.tokenizer_class(self.vocab_file)
vocab_size = len(tokenizer)
self.assertEqual(tokenizer.add_tokens(""), 0)
self.assertEqual(tokenizer.add_tokens("testoken"), 1)
self.assertEqual(tokenizer.add_tokens(["testoken1", "testtoken2"]), 2)
self.assertEqual(len(tokenizer), vocab_size + 3)
self.assertEqual(tokenizer.add_special_tokens({}), 0)
self.assertEqual(tokenizer.add_special_tokens({"bos_token": "[BOS]", "eos_token": "[EOS]"}), 2)
self.assertRaises(AssertionError, tokenizer.add_special_tokens, {"additional_special_tokens": "<testtoken1>"})
self.assertEqual(tokenizer.add_special_tokens({"additional_special_tokens": ["<testtoken2>"]}), 1)
self.assertEqual(
tokenizer.add_special_tokens({"additional_special_tokens": ["<testtoken3>", "<testtoken4>"]}), 2
)
self.assertIn("<testtoken3>", tokenizer.special_tokens_map["additional_special_tokens"])
self.assertIsInstance(tokenizer.special_tokens_map["additional_special_tokens"], list)
self.assertGreaterEqual(len(tokenizer.special_tokens_map["additional_special_tokens"]), 2)
self.assertEqual(len(tokenizer), vocab_size + 8)