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

Restore mask_position argument name #1185

Merged
merged 1 commit into from
Jul 31, 2023
Merged

Restore mask_position argument name #1185

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
32 changes: 17 additions & 15 deletions keras_nlp/layers/modeling/masked_lm_head.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,7 +123,6 @@ def __init__(
self.layer_norm_epsilon = layer_norm_epsilon
self.kernel_initializer = keras.initializers.get(kernel_initializer)
self.bias_initializer = keras.initializers.get(bias_initializer)
self._built = False

if vocabulary_size is None and embedding_weights is None:
raise ValueError(
Expand All @@ -142,7 +141,7 @@ def __init__(
)
self.vocabulary_size = shape[0]

def build(self, inputs_shape, masked_positions_shape=None):
def build(self, inputs_shape):
if self.embedding_weights is not None:
feature_size = self.embedding_weights.shape[-1]
else:
Expand All @@ -157,12 +156,13 @@ def build(self, inputs_shape, masked_positions_shape=None):
self._layer_norm = keras.layers.LayerNormalization(
epsilon=self.layer_norm_epsilon,
)
if masked_positions_shape:
gather_length = masked_positions_shape[1]
shape = (inputs_shape[0], gather_length, inputs_shape[-1])
self._dense.build(shape)
shape = (inputs_shape[0], gather_length, feature_size)
self._layer_norm.build(shape)
# The gather length does not affect any of our built variables, so
mattdangerw marked this conversation as resolved.
Show resolved Hide resolved
# we can pass any value here.
gather_length = None
shape = (inputs_shape[0], gather_length, inputs_shape[-1])
self._dense.build(shape)
shape = (inputs_shape[0], gather_length, feature_size)
self._layer_norm.build(shape)
if self.embedding_weights is None:
self._kernel = self.add_weight(
name="output_kernel",
Expand All @@ -177,10 +177,10 @@ def build(self, inputs_shape, masked_positions_shape=None):
dtype=self.dtype,
)

def call(self, inputs, masked_positions):
def call(self, inputs, mask_positions):
# Gather the encoded tokens at the masked indices.
masked_positions = ops.expand_dims(masked_positions, axis=-1)
x = ops.take_along_axis(inputs, masked_positions, axis=1)
mask_positions = ops.expand_dims(mask_positions, axis=-1)
x = ops.take_along_axis(inputs, mask_positions, axis=1)

# Apply a trainable linear transformation and a layer norm.
x = self._dense(x)
Expand Down Expand Up @@ -221,7 +221,9 @@ def get_config(self):
)
return config

def compute_output_shape(self, inputs_shape, masked_positions_shape):
output_shape = list(masked_positions_shape)
output_shape[-1] = self.vocabulary_size
return tuple(output_shape)
# TODO: restore this after https://github.com/keras-team/keras-core/pull/632
# is in a release!
# def compute_output_shape(self, inputs_shape, mask_positions_shape):
# output_shape = list(mask_positions_shape)
# output_shape[-1] = self.vocabulary_size
# return tuple(output_shape)
8 changes: 4 additions & 4 deletions keras_nlp/layers/modeling/masked_lm_head_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@ def test_valid_call(self):
)
encoded_tokens = keras.Input(shape=(10, 16))
positions = keras.Input(shape=(5,), dtype="int32")
outputs = head(encoded_tokens, masked_positions=positions)
outputs = head(encoded_tokens, mask_positions=positions)
model = keras.Model((encoded_tokens, positions), outputs)

token_data = ops.random.uniform(shape=(4, 10, 16))
Expand All @@ -48,7 +48,7 @@ def test_valid_call_with_embedding_weights(self):
# need to support this in the layer.
sequence = keras.Input(shape=(10, 32))
positions = keras.Input(shape=(5,), dtype="int32")
outputs = head(sequence, masked_positions=positions)
outputs = head(sequence, mask_positions=positions)
model = keras.Model((sequence, positions), outputs)
sequence_data = ops.random.uniform(shape=(4, 10, 32))
position_data = ops.random.randint(minval=0, maxval=10, shape=(4, 5))
Expand Down Expand Up @@ -106,7 +106,7 @@ def test_one_train_step(self):
)
encoded_tokens = keras.Input(shape=(10, 16))
positions = keras.Input(shape=(5,), dtype="int32")
outputs = head(encoded_tokens, masked_positions=positions)
outputs = head(encoded_tokens, mask_positions=positions)
model = keras.Model((encoded_tokens, positions), outputs)

token_data = ops.random.uniform(shape=(4, 10, 16))
Expand All @@ -126,7 +126,7 @@ def test_saved_model(self):
)
encoded_tokens = keras.Input(shape=(10, 16))
positions = keras.Input(shape=(5,), dtype="int32")
outputs = head(encoded_tokens, masked_positions=positions)
outputs = head(encoded_tokens, mask_positions=positions)
model = keras.Model((encoded_tokens, positions), outputs)

token_data = ops.random.uniform(shape=(4, 10, 16))
Expand Down