Added Kernel and Bias Initializers to Encoder and Decoder (#50)

* Added Kernel and Bias Initializer to
decoder

* Added Initializer to Encoder and Decoder

* Added Initializers to Expected Test Config

* Added Serialized Version to Config, Added New Test

* Fixed Docstring for Encoder and Decoder

* Changed initializer import to keras.initializer

* Removed Redudant Test From Encoder and Decoder

* Changed Default to Glorot Uniform and Zeros

* Ensure friendly error if bad arg on layer creation

* Ran Black Formatter

* Fixed Serialization Bug and Reran Black

* Added Additional Tests for Testing Value Error

* Keeping Attribute Set. From Const. Arg. Together

* New test for Value Error if Invalid Initializer

* Ran format and lint

* Fixed typo and also lines exceeding max length

Author: aflah02

keras_nlp/layers/transformer_decoder.py

@@ -39,13 +39,19 @@ class TransformerDecoder(keras.layers.Layer):
             activation function of feedforward network.
         layer_norm_epsilon: float, defaults to 1e-5. The eps value in layer
             normalization components.
+        kernel_initializer: string or tf.keras.initializers initializer,
+            defaults to "glorot_uniform". The kernel initializer for
+            the dense and multiheaded attention layers.
+        bias_initializer: string or tf.keras.initializers initializer,
+            defaults to "zeros". The bias initializer for
+            the dense and multiheaded attention layers.
         name: string, defaults to None. The name of the layer.
         **kwargs: other keyword arguments.
 
     Examples:
     ```python
     # Create a single transformer decoder layer.
-    decoder = keras_nlp.layer.TransformerDecoder(
+    decoder = keras_nlp.layers.TransformerDecoder(
         intermediate_dim=64, num_heads=8)
 
     # Create a simple model containing the decoder.
@@ -74,15 +80,19 @@ def __init__(
         dropout=0,
         activation="relu",
         layer_norm_epsilon=1e-05,
+        kernel_initializer="glorot_uniform",
+        bias_initializer="zeros",
         name=None,
         **kwargs,
     ):
         super().__init__(name=name, **kwargs)
         self.intermediate_dim = intermediate_dim
         self.num_heads = num_heads
         self.dropout = dropout
-        self.activation = activation
+        self.activation = keras.activations.get(activation)
         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
 
     def _build(self, input_shape):
@@ -95,12 +105,16 @@ def _build(self, input_shape):
             key_dim=self._attention_head_size,
             value_dim=self._attention_head_size,
             dropout=self.dropout,
+            kernel_initializer=self.kernel_initializer,
+            bias_initializer=self.bias_initializer,
         )
         self._encoder_decoder_attention_layer = keras.layers.MultiHeadAttention(
             num_heads=self.num_heads,
             key_dim=self._attention_head_size,
             value_dim=feature_size,
             dropout=self.dropout,
+            kernel_initializer=self.kernel_initializer,
+            bias_initializer=self.bias_initializer,
         )
 
         self._decoder_attention_layernorm = keras.layers.LayerNormalization()
@@ -114,11 +128,18 @@ def _build(self, input_shape):
         # First dense layer in the feedforward network, which maps input
         # feauture size to dimension `self.intermediate_dim`.
         self._intermediate_dense = keras.layers.Dense(
-            self.intermediate_dim, activation=self.activation
+            self.intermediate_dim,
+            activation=self.activation,
+            kernel_initializer=self.kernel_initializer,
+            bias_initializer=self.bias_initializer,
         )
         # Second dense layer in the feedforward network, which maps input
         # feature size back to the input feature size.
-        self._output_dense = keras.layers.Dense(feature_size)
+        self._output_dense = keras.layers.Dense(
+            feature_size,
+            kernel_initializer=self.kernel_initializer,
+            bias_initializer=self.bias_initializer,
+        )
         self._outputdropout = keras.layers.Dropout(rate=self.dropout)
 
     def _add_and_norm(self, input1, input2, norm_layer):
@@ -219,8 +240,14 @@ def get_config(self):
                 "intermediate_dim": self.intermediate_dim,
                 "num_heads": self.num_heads,
                 "dropout": self.dropout,
-                "activation": self.activation,
+                "activation": keras.activations.serialize(self.activation),
                 "layer_norm_epsilon": self.layer_norm_epsilon,
+                "kernel_initializer": keras.initializers.serialize(
+                    self.kernel_initializer
+                ),
+                "bias_initializer": keras.initializers.serialize(
+                    self.bias_initializer
+                ),
             }
         )
         return config

keras_nlp/layers/transformer_decoder_test.py

@@ -60,24 +60,46 @@ def test_get_config_and_from_config(self):
         decoder = transformer_decoder.TransformerDecoder(
             intermediate_dim=4,
             num_heads=2,
+            kernel_initializer="HeNormal",
+            bias_initializer="Zeros",
         )
+
         config = decoder.get_config()
+
         expected_config_subset = {
             "intermediate_dim": 4,
             "num_heads": 2,
             "dropout": 0,
             "activation": "relu",
             "layer_norm_epsilon": 1e-05,
+            "kernel_initializer": keras.initializers.serialize(
+                keras.initializers.HeNormal()
+            ),
+            "bias_initializer": keras.initializers.serialize(
+                keras.initializers.Zeros()
+            ),
         }
+
+        self.assertEqual(config, {**config, **expected_config_subset})
         self.assertEqual(config, {**config, **expected_config_subset})
 
         restored_decoder = transformer_decoder.TransformerDecoder.from_config(
             config,
         )
+
         self.assertEqual(
             restored_decoder.get_config(), {**config, **expected_config_subset}
         )
 
+    def test_value_error_when_invalid_kernel_inititalizer(self):
+        with self.assertRaises(ValueError):
+            transformer_decoder.TransformerDecoder(
+                intermediate_dim=4,
+                num_heads=2,
+                dropout=0.5,
+                kernel_initializer="Invalid",
+            )
+
     def test_one_training_step_of_transformer_encoder(self):
         class MyModel(keras.Model):
             def __init__(self):

keras_nlp/layers/transformer_encoder.py

@@ -37,14 +37,20 @@ class TransformerEncoder(keras.layers.Layer):
             activation function of feedforward network.
         layer_norm_epsilon: float, defaults to 1e-5. The epsilon value in layer
             normalization components.
+        kernel_initializer: string or tf.keras.initializers initializer,
+            defaults to "glorot_uniform". The kernel initializer for
+            the dense and multiheaded attention layers.
+        bias_initializer: string or tf.keras.initializers initializer,
+            defaults to "zeros". The bias initializer for
+            the dense and multiheaded attention layers.
         name: string, defaults to None. The name of the layer.
         **kwargs: other keyword arguments.
 
     Examples:
 
     ```python
     # Create a single transformer encoder layer.
-    encoder = keras_nlp.layer.TransformerEncoder(
+    encoder = keras_nlp.layers.TransformerEncoder(
         intermediate_dim=64, num_heads=8)
 
     # Create a simple model containing the encoder.
@@ -69,15 +75,19 @@ def __init__(
         dropout=0,
         activation="relu",
         layer_norm_epsilon=1e-05,
+        kernel_initializer="glorot_uniform",
+        bias_initializer="zeros",
         name=None,
         **kwargs
     ):
         super().__init__(name=name, **kwargs)
         self.intermediate_dim = intermediate_dim
         self.num_heads = num_heads
         self.dropout = dropout
-        self.activation = activation
+        self.activation = keras.activations.get(activation)
         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
 
     def _build(self, input_shape):
@@ -90,6 +100,8 @@ def _build(self, input_shape):
             key_dim=self._attention_head_size,
             value_dim=self._attention_head_size,
             dropout=self.dropout,
+            kernel_initializer=self.kernel_initializer,
+            bias_initializer=self.bias_initializer,
         )
 
         self._attention_layernorm = keras.layers.LayerNormalization()
@@ -98,9 +110,16 @@ def _build(self, input_shape):
         self._attentiondropout = keras.layers.Dropout(rate=self.dropout)
 
         self._intermediate_dense = keras.layers.Dense(
-            self.intermediate_dim, activation=self.activation
+            self.intermediate_dim,
+            activation=self.activation,
+            kernel_initializer=self.kernel_initializer,
+            bias_initializer=self.bias_initializer,
+        )
+        self._output_dense = keras.layers.Dense(
+            feature_size,
+            kernel_initializer=self.kernel_initializer,
+            bias_initializer=self.bias_initializer,
         )
-        self._output_dense = keras.layers.Dense(feature_size)
         self._outputdropout = keras.layers.Dropout(rate=self.dropout)
 
     def _add_and_norm(self, input1, input2, norm_layer):
@@ -161,8 +180,14 @@ def get_config(self):
                 "intermediate_dim": self.intermediate_dim,
                 "num_heads": self.num_heads,
                 "dropout": self.dropout,
-                "activation": self.activation,
+                "activation": keras.activations.serialize(self.activation),
                 "layer_norm_epsilon": self.layer_norm_epsilon,
+                "kernel_initializer": keras.initializers.serialize(
+                    self.kernel_initializer
+                ),
+                "bias_initializer": keras.initializers.serialize(
+                    self.bias_initializer
+                ),
             }
         )
         return config

keras_nlp/layers/transformer_encoder_test.py

@@ -50,24 +50,45 @@ def test_get_config_and_from_config(self):
         encoder = transformer_encoder.TransformerEncoder(
             intermediate_dim=4,
             num_heads=2,
+            kernel_initializer="HeNormal",
+            bias_initializer="Zeros",
         )
+
         config = encoder.get_config()
+
         expected_config_subset = {
             "intermediate_dim": 4,
             "num_heads": 2,
             "dropout": 0,
             "activation": "relu",
             "layer_norm_epsilon": 1e-05,
+            "kernel_initializer": keras.initializers.serialize(
+                keras.initializers.HeNormal()
+            ),
+            "bias_initializer": keras.initializers.serialize(
+                keras.initializers.Zeros()
+            ),
         }
+
         self.assertEqual(config, {**config, **expected_config_subset})
 
         restored_encoder = transformer_encoder.TransformerEncoder.from_config(
             config,
         )
+
         self.assertEqual(
             restored_encoder.get_config(), {**config, **expected_config_subset}
         )
 
+    def test_value_error_when_invalid_kernel_inititalizer(self):
+        with self.assertRaises(ValueError):
+            transformer_encoder.TransformerEncoder(
+                intermediate_dim=4,
+                num_heads=2,
+                dropout=0.5,
+                kernel_initializer="Invalid",
+            )
+
     def test_one_training_step_of_transformer_encoder(self):
         encoder = transformer_encoder.TransformerEncoder(
             intermediate_dim=4,