Adding [T5/MT5/UMT5]ForTokenClassification (#28443)
* Adding [T5/MT5/UMT5]ForTokenClassification * Add auto mappings for T5ForTokenClassification and variants * Adding ForTokenClassification to the list of models * Adding attention_mask param to the T5ForTokenClassification test * Remove outdated comment in test * Adding EncoderOnly and Token Classification tests for MT5 and UMT5 * Fix typo in umt5 string * Add tests for all the existing MT5 models * Fix wrong comment in dependency_versions_table * Reverting change to common test for _keys_to_ignore_on_load_missing The test is correctly picking up redundant keys in _keys_to_ignore_on_load_missing. * Removing _keys_to_ignore_on_missing from MT5 since the key is not used in the model * Add fix-copies to MT5ModelTest
This commit is contained in:
JB (Don)
@@ -18,7 +18,7 @@ import pickle
|
||||
import tempfile
|
||||
import unittest
|
||||
|
||||
from transformers import T5Config, is_torch_available
|
||||
from transformers import UMT5Config, is_torch_available
|
||||
from transformers.models.auto.modeling_auto import MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES
|
||||
from transformers.testing_utils import (
|
||||
require_sentencepiece,
|
||||
@@ -30,6 +30,7 @@ from transformers.testing_utils import (
|
||||
from transformers.utils import is_torch_fx_available
|
||||
|
||||
from ...generation.test_utils import GenerationTesterMixin
|
||||
from ...test_configuration_common import ConfigTester
|
||||
from ...test_modeling_common import ModelTesterMixin, _config_zero_init, ids_tensor
|
||||
from ...test_pipeline_mixin import PipelineTesterMixin
|
||||
|
||||
@@ -43,9 +44,11 @@ if is_torch_available():
|
||||
|
||||
from transformers import (
|
||||
AutoTokenizer,
|
||||
UMT5EncoderModel,
|
||||
UMT5ForConditionalGeneration,
|
||||
UMT5ForQuestionAnswering,
|
||||
UMT5ForSequenceClassification,
|
||||
UMT5ForTokenClassification,
|
||||
UMT5Model,
|
||||
)
|
||||
|
||||
@@ -100,7 +103,7 @@ class UMT5ModelTester:
|
||||
self.decoder_layers = decoder_layers
|
||||
|
||||
def get_large_model_config(self):
|
||||
return T5Config.from_pretrained("google/umt5-base")
|
||||
return UMT5Config.from_pretrained("google/umt5-base")
|
||||
|
||||
def prepare_inputs_dict(
|
||||
self,
|
||||
@@ -160,7 +163,7 @@ class UMT5ModelTester:
|
||||
return config, inputs_dict
|
||||
|
||||
def get_pipeline_config(self):
|
||||
return T5Config(
|
||||
return UMT5Config(
|
||||
vocab_size=166, # t5 forces 100 extra tokens
|
||||
d_model=self.hidden_size,
|
||||
d_ff=self.d_ff,
|
||||
@@ -178,7 +181,7 @@ class UMT5ModelTester:
|
||||
)
|
||||
|
||||
def get_config(self):
|
||||
return T5Config(
|
||||
return UMT5Config(
|
||||
vocab_size=self.vocab_size,
|
||||
d_model=self.hidden_size,
|
||||
d_ff=self.d_ff,
|
||||
@@ -556,6 +559,176 @@ class UMT5ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin
|
||||
pass
|
||||
|
||||
|
||||
# Copied from tests.models.t5.test_modeling_t5.T5EncoderOnlyModelTester with T5->UMT5
|
||||
class UMT5EncoderOnlyModelTester:
|
||||
def __init__(
|
||||
self,
|
||||
parent,
|
||||
vocab_size=99,
|
||||
batch_size=13,
|
||||
encoder_seq_length=7,
|
||||
# For common tests
|
||||
use_attention_mask=True,
|
||||
hidden_size=32,
|
||||
num_hidden_layers=2,
|
||||
num_attention_heads=4,
|
||||
d_ff=37,
|
||||
relative_attention_num_buckets=8,
|
||||
is_training=False,
|
||||
dropout_rate=0.1,
|
||||
initializer_factor=0.002,
|
||||
is_encoder_decoder=False,
|
||||
eos_token_id=1,
|
||||
pad_token_id=0,
|
||||
scope=None,
|
||||
):
|
||||
self.parent = parent
|
||||
self.batch_size = batch_size
|
||||
self.encoder_seq_length = encoder_seq_length
|
||||
# For common tests
|
||||
self.seq_length = self.encoder_seq_length
|
||||
self.use_attention_mask = use_attention_mask
|
||||
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.d_ff = d_ff
|
||||
self.relative_attention_num_buckets = relative_attention_num_buckets
|
||||
self.dropout_rate = dropout_rate
|
||||
self.initializer_factor = initializer_factor
|
||||
self.eos_token_id = eos_token_id
|
||||
self.pad_token_id = pad_token_id
|
||||
self.is_encoder_decoder = is_encoder_decoder
|
||||
self.scope = None
|
||||
self.is_training = is_training
|
||||
|
||||
def get_large_model_config(self):
|
||||
return UMT5Config.from_pretrained("t5-base")
|
||||
|
||||
def prepare_config_and_inputs(self):
|
||||
input_ids = ids_tensor([self.batch_size, self.encoder_seq_length], self.vocab_size)
|
||||
|
||||
attention_mask = None
|
||||
if self.use_attention_mask:
|
||||
attention_mask = ids_tensor([self.batch_size, self.encoder_seq_length], vocab_size=2)
|
||||
|
||||
config = UMT5Config(
|
||||
vocab_size=self.vocab_size,
|
||||
d_model=self.hidden_size,
|
||||
d_ff=self.d_ff,
|
||||
d_kv=self.hidden_size // self.num_attention_heads,
|
||||
num_layers=self.num_hidden_layers,
|
||||
num_heads=self.num_attention_heads,
|
||||
relative_attention_num_buckets=self.relative_attention_num_buckets,
|
||||
dropout_rate=self.dropout_rate,
|
||||
initializer_factor=self.initializer_factor,
|
||||
eos_token_id=self.eos_token_id,
|
||||
bos_token_id=self.pad_token_id,
|
||||
pad_token_id=self.pad_token_id,
|
||||
is_encoder_decoder=self.is_encoder_decoder,
|
||||
)
|
||||
|
||||
return (
|
||||
config,
|
||||
input_ids,
|
||||
attention_mask,
|
||||
)
|
||||
|
||||
def create_and_check_model(
|
||||
self,
|
||||
config,
|
||||
input_ids,
|
||||
attention_mask,
|
||||
):
|
||||
model = UMT5EncoderModel(config=config)
|
||||
model.to(torch_device)
|
||||
model.eval()
|
||||
result = model(
|
||||
input_ids=input_ids,
|
||||
attention_mask=attention_mask,
|
||||
)
|
||||
result = model(input_ids=input_ids)
|
||||
encoder_output = result.last_hidden_state
|
||||
|
||||
self.parent.assertEqual(encoder_output.size(), (self.batch_size, self.encoder_seq_length, self.hidden_size))
|
||||
|
||||
def create_and_check_model_fp16_forward(
|
||||
self,
|
||||
config,
|
||||
input_ids,
|
||||
attention_mask,
|
||||
):
|
||||
model = UMT5EncoderModel(config=config).to(torch_device).half().eval()
|
||||
output = model(input_ids, attention_mask=attention_mask)["last_hidden_state"]
|
||||
self.parent.assertFalse(torch.isnan(output).any().item())
|
||||
|
||||
def create_and_check_with_token_classification_head(
|
||||
self,
|
||||
config,
|
||||
input_ids,
|
||||
attention_mask,
|
||||
):
|
||||
labels = torch.tensor([1] * self.seq_length * self.batch_size, dtype=torch.long, device=torch_device)
|
||||
model = UMT5ForTokenClassification(config=config).to(torch_device).eval()
|
||||
outputs = model(
|
||||
input_ids=input_ids,
|
||||
labels=labels,
|
||||
attention_mask=attention_mask,
|
||||
)
|
||||
self.parent.assertEqual(outputs["logits"].size(), (self.batch_size, self.seq_length, config.num_labels))
|
||||
self.parent.assertEqual(outputs["loss"].size(), ())
|
||||
|
||||
def prepare_config_and_inputs_for_common(self):
|
||||
config_and_inputs = self.prepare_config_and_inputs()
|
||||
(
|
||||
config,
|
||||
input_ids,
|
||||
attention_mask,
|
||||
) = config_and_inputs
|
||||
|
||||
inputs_dict = {
|
||||
"input_ids": input_ids,
|
||||
"attention_mask": attention_mask,
|
||||
}
|
||||
return config, inputs_dict
|
||||
|
||||
|
||||
# Copied from tests.models.t5.test_modeling_t5.T5EncoderOnlyModelTest with T5->UMT5
|
||||
class UMT5EncoderOnlyModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
|
||||
all_model_classes = (UMT5EncoderModel, UMT5ForTokenClassification) if is_torch_available() else ()
|
||||
test_pruning = False
|
||||
test_resize_embeddings = False
|
||||
test_model_parallel = True
|
||||
pipeline_model_mapping = (
|
||||
{
|
||||
"token-classification": UMT5ForTokenClassification,
|
||||
}
|
||||
if is_torch_available()
|
||||
else {}
|
||||
)
|
||||
all_parallelizable_model_classes = (UMT5EncoderModel,) if is_torch_available() else ()
|
||||
|
||||
def setUp(self):
|
||||
self.model_tester = UMT5EncoderOnlyModelTester(self)
|
||||
self.config_tester = ConfigTester(self, config_class=UMT5Config, d_model=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)
|
||||
|
||||
@unittest.skipIf(torch_device == "cpu", "Cant do half precision")
|
||||
def test_model_fp16_forward(self):
|
||||
config_and_inputs = self.model_tester.prepare_config_and_inputs()
|
||||
self.model_tester.create_and_check_model_fp16_forward(*config_and_inputs)
|
||||
|
||||
def test_with_token_classification_head(self):
|
||||
config_and_inputs = self.model_tester.prepare_config_and_inputs()
|
||||
self.model_tester.create_and_check_with_token_classification_head(*config_and_inputs)
|
||||
|
||||
|
||||
@require_torch
|
||||
@require_sentencepiece
|
||||
@require_tokenizers
|
||||
|
||||
Reference in New Issue
Block a user