Skip to content

Speeding the tests for whisper #887

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

Merged
merged 2 commits into from
Mar 22, 2023
Merged

Speeding the tests for whisper #887

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d9ad1ac
modified tests for whisper
susnato Mar 20, 2023
381d6f8
comments
susnato 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
105 changes: 45 additions & 60 deletions keras_nlp/models/whisper/whisper_backbone_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,89 +26,73 @@

class WhisperBackboneTest(tf.test.TestCase, parameterized.TestCase):
def setUp(self):
self.model = WhisperBackbone(
vocabulary_size=1000,
self.backbone = WhisperBackbone(
vocabulary_size=10,
num_layers=2,
num_heads=2,
hidden_dim=64,
intermediate_dim=128,
max_encoder_sequence_length=128,
max_decoder_sequence_length=96,
hidden_dim=2,
intermediate_dim=4,
max_encoder_sequence_length=6,
max_decoder_sequence_length=6,
)
self.batch_size = 8
self.input_batch = {
"encoder_features": tf.ones(
(
self.batch_size,
self.model.max_encoder_sequence_length,
NUM_MELS,
),
dtype="int32",
),
"decoder_token_ids": tf.ones(
(self.batch_size, self.model.max_decoder_sequence_length),
dtype="int32",
),
"decoder_padding_mask": tf.ones(
(self.batch_size, self.model.max_decoder_sequence_length),
dtype="int32",
),
"encoder_features": tf.ones((2, 5, NUM_MELS), dtype="int32"),
"decoder_token_ids": tf.ones((2, 5), dtype="int32"),
"decoder_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_whisper(self):
self.model(self.input_batch)
self.backbone(self.input_batch)

def test_token_embedding(self):
output = self.backbone.token_embedding(
self.input_batch["decoder_token_ids"]
)
self.assertEqual(output.shape, (2, 5, 2))

def test_name(self):
# Check default name passed through
self.assertRegexpMatches(self.model.name, "whisper_backbone")
self.assertRegexpMatches(self.backbone.name, "whisper_backbone")

def test_variable_sequence_length_call_whisper(self):
for seq_length in (25, 50, 75):
for seq_length in (2, 3, 4):
input_data = {
"encoder_features": tf.ones(
(self.batch_size, seq_length, NUM_MELS),
dtype="int32",
),
"decoder_token_ids": tf.ones(
(self.batch_size, seq_length), dtype="int32"
),
"decoder_padding_mask": tf.ones(
(self.batch_size, seq_length), dtype="int32"
(2, seq_length, NUM_MELS), dtype="int32"
),
"decoder_token_ids": tf.ones((2, seq_length), dtype="int32"),
"decoder_padding_mask": tf.ones((2, seq_length), dtype="int32"),
}
self.model(input_data)
self.backbone(input_data)

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

@parameterized.named_parameters(
("jit_compile_false", False), ("jit_compile_true", True)
)
def test_compile_batched_ds(self, jit_compile):
self.model.compile(jit_compile=jit_compile)
self.model.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_key_projection_bias_absence(self):
# Check only for the first encoder layer and first decoder layer.
self.assertIsNone(
self.model.get_layer(
self.backbone.get_layer(
"transformer_encoder_layer_0"
)._self_attention_layer._key_dense.bias
)
self.assertIsNone(
self.model.get_layer(
self.backbone.get_layer(
"transformer_decoder_layer_0"
)._self_attention_layer._key_dense.bias
)
self.assertIsNone(
self.model.get_layer(
self.backbone.get_layer(
"transformer_decoder_layer_0"
)._cross_attention_layer._key_dense.bias
)
Expand All @@ -117,10 +101,11 @@ def test_key_projection_bias_absence(self):
("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 @@ -143,14 +128,14 @@ def test_saved_model(self, save_format, filename):
class WhisperBackboneTPUTest(tf.test.TestCase, parameterized.TestCase):
def setUp(self):
with self.tpu_strategy.scope():
self.model = WhisperBackbone(
vocabulary_size=1000,
self.backbone = WhisperBackbone(
vocabulary_size=10,
num_layers=2,
num_heads=2,
hidden_dim=64,
intermediate_dim=128,
max_encoder_sequence_length=128,
max_decoder_sequence_length=64,
hidden_dim=2,
intermediate_dim=4,
max_encoder_sequence_length=6,
max_decoder_sequence_length=6,
)

self.input_batch = {
Expand All @@ -175,5 +160,5 @@ 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)