Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Speed up default RoBERTa testing roughly 3x #897

Merged
merged 3 commits into from
Mar 22, 2023
Merged

Speed up default RoBERTa testing roughly 3x #897

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f7ae534
initial commit for roberta test improvement :octocat:
shivance Mar 20, 2023
c5db160
fixing up errors; wrapping up
shivance Mar 21, 2023
50e276e
fixing save model test
shivance Mar 21, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
73 changes: 33 additions & 40 deletions keras_nlp/models/roberta/roberta_backbone_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,72 +26,65 @@

class RobertaBackboneTest(tf.test.TestCase, parameterized.TestCase):
def setUp(self):
self.model = RobertaBackbone(
vocabulary_size=1000,
self.backbone = RobertaBackbone(
vocabulary_size=10,
num_layers=2,
num_heads=2,
hidden_dim=64,
intermediate_dim=128,
max_sequence_length=128,
hidden_dim=2,
intermediate_dim=4,
max_sequence_length=5,
)
self.batch_size = 8
self.input_batch = {
"token_ids": tf.ones(
(self.batch_size, self.model.max_sequence_length), dtype="int32"
),
"padding_mask": tf.ones(
(self.batch_size, self.model.max_sequence_length), dtype="int32"
),
"token_ids": tf.ones((2, 5), dtype="int32"),
"padding_mask": tf.ones((2, 5), dtype="int32"),
}

self.input_dataset = tf.data.Dataset.from_tensor_slices(
self.input_batch
).batch(2)

def test_valid_call_roberta(self):
self.model(self.input_batch)
self.backbone(self.input_batch)

# Check default name passed through
self.assertRegexpMatches(self.model.name, "roberta_backbone")
def test_token_embedding(self):
output = self.backbone.token_embedding(self.input_batch["token_ids"])
self.assertEqual(output.shape, (2, 5, 2))

@parameterized.named_parameters(
("jit_compile_false", False), ("jit_compile_true", True)
)
def test_roberta_compile(self, jit_compile):
self.model.compile(jit_compile=jit_compile)
self.model.predict(self.input_batch)
def test_name(self):
self.assertRegexpMatches(self.backbone.name, "roberta_backbone")

@parameterized.named_parameters(
("jit_compile_false", False), ("jit_compile_true", True)
)
def test_roberta_compile_batched_ds(self, jit_compile):
self.model.compile(jit_compile=jit_compile)
self.model.predict(self.input_dataset)
def test_predict(self):
self.backbone.predict(self.input_batch)
self.backbone.predict(self.input_dataset)

def test_serialization(self):
new_backbone = keras.utils.deserialize_keras_object(
keras.utils.serialize_keras_object(self.backbone)
)
self.assertEqual(new_backbone.get_config(), self.backbone.get_config())

def test_variable_sequence_length_call_roberta(self):
for seq_length in (25, 50, 75):
for seq_length in (2, 3, 4):
input_data = {
"token_ids": tf.ones(
(self.batch_size, seq_length), dtype="int32"
),
"padding_mask": tf.ones(
(self.batch_size, seq_length), dtype="int32"
),
"token_ids": tf.ones((2, seq_length), dtype="int32"),
"padding_mask": tf.ones((2, seq_length), dtype="int32"),
}
output = self.model(input_data)
output = self.backbone(input_data)
self.assertAllEqual(
tf.shape(output),
[self.batch_size, seq_length, self.model.hidden_dim],
[2, seq_length, self.backbone.hidden_dim],
)

@parameterized.named_parameters(
("tf_format", "tf", "model"),
("keras_format", "keras_v3", "model.keras"),
)
@pytest.mark.large # Saving is slow, so mark these large.
def test_saved_model(self, save_format, filename):
model_output = self.model(self.input_batch)
model_output = self.backbone(self.input_batch)
save_path = os.path.join(self.get_temp_dir(), filename)
self.model.save(save_path, save_format=save_format)
self.backbone.save(save_path, save_format=save_format)
restored_model = keras.models.load_model(save_path)

# Check we got the real object back.
Expand All @@ -107,7 +100,7 @@ def test_saved_model(self, save_format, filename):
class RobertaBackboneTPUTest(tf.test.TestCase, parameterized.TestCase):
def setUp(self):
with self.tpu_strategy.scope():
self.model = RobertaBackbone(
self.backbone = RobertaBackbone(
vocabulary_size=1000,
num_layers=2,
num_heads=2,
Expand All @@ -124,5 +117,5 @@ def setUp(self):
).batch(2)

def test_predict(self):
self.model.compile()
self.model.predict(self.input_dataset)
self.backbone.compile()
self.backbone.predict(self.input_dataset)
77 changes: 31 additions & 46 deletions keras_nlp/models/roberta/roberta_classifier_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@

import os

import pytest
import tensorflow as tf
from absl.testing import parameterized
from tensorflow import keras
Expand All @@ -27,14 +28,6 @@

class RobertaClassifierTest(tf.test.TestCase, parameterized.TestCase):
def setUp(self):
self.backbone = RobertaBackbone(
vocabulary_size=1000,
num_layers=2,
num_heads=2,
hidden_dim=64,
intermediate_dim=128,
max_sequence_length=128,
)
self.vocab = {
"<s>": 0,
"<pad>": 1,
Expand All @@ -58,77 +51,69 @@ def setUp(self):
self.merges = merges
self.preprocessor = RobertaPreprocessor(
RobertaTokenizer(vocabulary=self.vocab, merges=self.merges),
sequence_length=8,
sequence_length=5,
)
self.backbone = RobertaBackbone(
vocabulary_size=self.preprocessor.tokenizer.vocabulary_size(),
num_layers=2,
num_heads=2,
hidden_dim=2,
intermediate_dim=4,
max_sequence_length=self.preprocessor.packer.sequence_length,
)
self.classifier = RobertaClassifier(
self.backbone,
4,
preprocessor=self.preprocessor,
)
self.classifier_no_preprocessing = RobertaClassifier(
self.backbone,
4,
preprocessor=None,
)

# Setup data.
self.raw_batch = tf.constant(
[
" airplane at airport",
" the airplane is the best",
" the best airport",
" kohli is the best",
]
)
self.preprocessed_batch = self.preprocessor(self.raw_batch)
self.raw_dataset = tf.data.Dataset.from_tensor_slices(
(self.raw_batch, tf.ones((4,)))
(self.raw_batch, tf.ones((2,)))
).batch(2)
self.preprocessed_dataset = self.raw_dataset.map(self.preprocessor)

def test_valid_call_classifier(self):
self.classifier(self.preprocessed_batch)

@parameterized.named_parameters(
("jit_compile_false", False), ("jit_compile_true", True)
)
def test_roberta_classifier_predict(self, jit_compile):
self.classifier.compile(jit_compile=jit_compile)
def test_classifier_predict(self):
self.classifier.predict(self.raw_batch)
self.classifier.preprocessor = None
self.classifier.predict(self.preprocessed_batch)

@parameterized.named_parameters(
("jit_compile_false", False), ("jit_compile_true", True)
)
def test_roberta_classifier_predict_no_preprocessing(self, jit_compile):
self.classifier_no_preprocessing.compile(jit_compile=jit_compile)
self.classifier_no_preprocessing.predict(self.preprocessed_batch)

def test_roberta_classifier_fit_default_compile(self):
def test_classifier_fit(self):
self.classifier.fit(self.raw_dataset)
self.classifier.preprocessor = None
self.classifier.fit(self.preprocessed_dataset)

@parameterized.named_parameters(
("jit_compile_false", False), ("jit_compile_true", True)
)
def test_roberta_classifier_fit(self, jit_compile):
def test_classifier_fit_no_xla(self):
self.classifier.preprocessor = None
self.classifier.compile(
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
jit_compile=jit_compile,
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=False),
jit_compile=False,
)
self.classifier.fit(self.raw_dataset)

@parameterized.named_parameters(
("jit_compile_false", False), ("jit_compile_true", True)
)
def test_roberta_classifier_fit_no_preprocessing(self, jit_compile):
self.classifier_no_preprocessing.compile(
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
jit_compile=jit_compile,
self.classifier.fit(self.preprocessed_dataset)

def test_serialization(self):
config = keras.utils.serialize_keras_object(self.classifier)
new_classifier = keras.utils.deserialize_keras_object(config)
self.assertEqual(
new_classifier.get_config(),
self.classifier.get_config(),
)
self.classifier_no_preprocessing.fit(self.preprocessed_dataset)

@parameterized.named_parameters(
("tf_format", "tf", "model"),
("keras_format", "keras_v3", "model.keras"),
)
@pytest.mark.large # Saving is slow, so mark these large.
def test_saved_model(self, save_format, filename):
model_output = self.classifier.predict(self.raw_batch)
save_path = os.path.join(self.get_temp_dir(), filename)
Expand Down
10 changes: 10 additions & 0 deletions keras_nlp/models/roberta/roberta_masked_lm_preprocessor_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@

import os

import pytest
import tensorflow as tf
from absl.testing import parameterized
from tensorflow import keras
Expand Down Expand Up @@ -140,10 +141,19 @@ def test_no_masking_zero_rate(self):
self.assertAllEqual(y, [0, 0, 0, 0, 0])
self.assertAllEqual(sw, [0.0, 0.0, 0.0, 0.0, 0.0])

def test_serialization(self):
config = keras.utils.serialize_keras_object(self.preprocessor)
new_preprocessor = keras.utils.deserialize_keras_object(config)
self.assertEqual(
new_preprocessor.get_config(),
self.preprocessor.get_config(),
)

@parameterized.named_parameters(
("tf_format", "tf", "model"),
("keras_format", "keras_v3", "model.keras"),
)
@pytest.mark.large # Saving is slow, so mark these large.
def test_saved_model(self, save_format, filename):
input_data = tf.constant([" airplane at airport"])

Expand Down
Loading