Adding GPTNeoXBackbone

keras_nlp/layers/masked_lm_mask_generator.py

@@ -174,11 +174,7 @@ def call(self, inputs):
             # convert dense to ragged.
             inputs = tf.RaggedTensor.from_tensor(inputs)
 
-        (
-            token_ids,
-            mask_positions,
-            mask_ids,
-        ) = tf_text.mask_language_model(
+        (token_ids, mask_positions, mask_ids,) = tf_text.mask_language_model(
             inputs,
             item_selector=self._random_selector,
             mask_values_chooser=self._mask_values_chooser,

keras_nlp/models/gpt_neox/gpt_neox_attention.py

@@ -0,0 +1,140 @@
+# Copyright 2023 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import tensorflow as tf
+from tensorflow import keras
+
+from keras_nlp.layers.transformer_layer_utils import compute_causal_mask
+from keras_nlp.models.gpt_neox.rotary_embedding import RotaryEmbedding
+
+
+class GPTNeoXAttention(keras.layers.Layer):
+    def __init__(
+        self,
+        num_heads,
+        hidden_dim,
+        rotary_pct=0.25,
+        max_position_embeddings=2048,
+    ):
+
+        super().__init__()
+        self.num_heads = num_heads
+        self.hidden_dim = hidden_dim
+        self.head_dim = hidden_dim // num_heads
+        self.rotary_dim = self.head_dim * rotary_pct
+        self.max_position_embeddings = max_position_embeddings
+        self.rotary_embedding = RotaryEmbedding(self.rotary_pct)
+        self.qkv = keras.layers.Dense(3 * self.hidden_dim)
+        self.dense = keras.layers.Dense(self.hidden_dim)
+
+    def _compute_attention(
+        self, query, key, value, attention_mask=None, head_mask=None
+    ):
+
+        batch_size, _, query_len, _ = tf.shape(query)
+        key_len = tf.shape(key)[-2]
+        # causal_mask = self.bias[:, :, key_len - query_len : key_len, :key_len]
+        causal_mask = compute_causal_mask(batch_size, key_len, key_len)
+
+        query = tf.reshape(
+            query, [batch_size * self.num_heads, query_len, self.head_dim]
+        )
+        key = tf.reshape(
+            key, [batch_size * self.num_heads, query_len, self.head_dim]
+        )
+        attention_scores = tf.zeros(
+            [batch_size * self.num_heads, query_len, self.head_dim],
+            dtype=query.dtype,
+        )
+
+        attention_scores = tf.linalg.matmul(
+            attention_scores,
+            query,
+            tf.transpose(key, perm=[0, 2, 1]),
+            beta=1.0,
+            alpha=(tf.constant(1.0)),
+        )
+        attention_scores = tf.reshape(
+            attention_scores, [batch_size, self.num_heads, query_len, key_len]
+        )
+        mask_value = tf.constant(float("-inf"), dtype=attention_scores.dtype)
+        attention_scores = tf.where(causal_mask, attention_scores, mask_value)
+
+        if attention_mask is not None:
+            attention_scores += attention_mask
+
+        attention_scores = tf.cast(
+            tf.nn.softmax(attention_scores, axis=-1), dtype=value.dtype
+        )
+
+        if head_mask is not None:
+            attention_scores *= head_mask
+
+        attention_output = tf.matmul(attention_scores, value)
+        return attention_output, attention_scores
+
+    def call(
+        self,
+        hidden_states,
+        attention_mask,
+        head_mask,
+        layer_past,
+        return_attention_scores,
+    ):
+
+        qkv = self.qkv(hidden_states)
+        new_qkv_shape = tf.shape(hidden_states)[:-1] + [
+            self.num_heads,
+            self.head_dim,
+        ]
+        qkv = tf.reshape(qkv, new_qkv_shape)
+
+        query = tf.transpose(qkv[..., : self.head_dim], (0, 2, 1, 3))
+        key = tf.transpose(
+            qkv[..., : self.head_dim : 2 * self.head_dim], (0, 2, 1, 3)
+        )
+        value = tf.transpose(qkv[..., self.head_dim :], (0, 2, 1, 3))
+
+        query_rot, query_pass = (
+            query[..., : self.rotary_dim],
+            query[..., self.rotary_dim :],
+        )
+        key_rot, key_pass = (
+            key[..., : self.rotary_dim],
+            key[..., self.rotary_dim :],
+        )
+
+        query, key = self.rotary_embedding(query_rot, key_rot)
+        query = tf.concat((query, query_pass), dim=-1)
+        key = tf.concat((key, key_pass), dim=-1)
+
+        if layer_past is not None:
+            past_key, past_value = layer_past
+            key = tf.concat((past_key, key), axis=-2)
+            value = tf.concat((past_value, value), axis=-2)
+
+        attention_output, attention_scores = self._compute_attention(
+            query, key, value, attention_mask, head_mask
+        )
+        new_shape = tf.shape(attention_output)[:-2] + (
+            self.num_heads * self.head_dim
+        )
+        attention_output = tf.reshape(
+            tf.transpose(attention_output, (0, 2, 1, 3)), new_shape
+        )
+        attention_output = self.dense(attention_output)
+
+        if return_attention_scores:
+            return (attention_output, attention_scores)
+
+        return attention_output

keras_nlp/models/gpt_neox/gpt_neox_backbone.py

@@ -0,0 +1,168 @@
+# Copyright 2023 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import tensorflow as tf
+from tensorflow import keras
+from tensorflow.experimental import dtensor
+from tensorflow.experimental.dtensor import Layout
+
+from keras_nlp.api_export import keras_nlp_export
+from keras_nlp.layers.position_embedding import PositionEmbedding
+from keras_nlp.models.backbone import Backbone
+from keras_nlp.models.gpt_neox.gpt_neox_decoder import GPTNeoXDecoder
+
+
+def _gpt_2_kernel_initializer(stddev=0.02):
+    return keras.initializers.RandomNormal(stddev=stddev)
+
+
+@keras_nlp_export("keras_nlp.models.GPT2Backbone")
+class GPTNeoXBackbone(Backbone):
+    def __init__(
+        self,
+        vocabulary_size,
+        num_layers,
+        num_heads,
+        hidden_dim,
+        intermediate_dim,
+        dropout=0.1,
+        max_sequence_length=1024,
+        **kwargs,
+    ):
+        # Inputs
+        token_ids = keras.Input(shape=(None,), dtype="int32", name="token_ids")
+        padding_mask = keras.Input(
+            shape=(None,), dtype="int32", name="padding_mask"
+        )
+
+        # Embed tokens, positions.
+        token_embedding = keras.layers.Embedding(
+            input_dim=vocabulary_size,
+            output_dim=hidden_dim,
+            embeddings_initializer=_gpt_2_kernel_initializer(stddev=0.01),
+            name="token_embedding",
+        )(token_ids)
+
+        # Can't use `TokenAndPositionEmbedding` layer here because of different
+        # initializers.
+        position_embedding = PositionEmbedding(
+            initializer=_gpt_2_kernel_initializer(stddev=0.02),
+            sequence_length=max_sequence_length,
+            name="position_embedding",
+        )(token_embedding)
+
+        # Sum and apply dropout to embeddings.
+        x = keras.layers.Add(name="embeddings_add")(
+            (token_embedding, position_embedding)
+        )
+        x = keras.layers.Dropout(
+            dropout,
+            name="embeddings_dropout",
+        )(x)
+
+        # Apply successive transformer decoder blocks.
+        for i in range(num_layers):
+            x = GPTNeoXDecoder(
+                intermediate_dim=intermediate_dim,
+                num_heads=num_heads,
+                dropout=dropout,
+                layer_norm_epsilon=1e-05,
+                activation=lambda x: keras.activations.gelu(
+                    x, approximate=True
+                ),
+                kernel_initializer=_gpt_2_kernel_initializer(stddev=0.02),
+                normalize_first=True,
+                name=f"transformer_layer_{i}",
+            )(x, decoder_padding_mask=padding_mask)
+
+        sequence_output = keras.layers.LayerNormalization(
+            name="layer_norm",
+            axis=-1,
+            epsilon=1e-05,
+            dtype=tf.float32,
+        )(x)
+
+        # Instantiate using Functional API Model constructor
+        super().__init__(
+            inputs={
+                "token_ids": token_ids,
+                "padding_mask": padding_mask,
+            },
+            outputs=sequence_output,
+            **kwargs,
+        )
+        # All references to `self` below this line
+        self.vocabulary_size = vocabulary_size
+        self.num_layers = num_layers
+        self.num_heads = num_heads
+        self.hidden_dim = hidden_dim
+        self.intermediate_dim = intermediate_dim
+        self.dropout = dropout
+        self.max_sequence_length = max_sequence_length
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "vocabulary_size": self.vocabulary_size,
+                "num_layers": self.num_layers,
+                "num_heads": self.num_heads,
+                "hidden_dim": self.hidden_dim,
+                "intermediate_dim": self.intermediate_dim,
+                "dropout": self.dropout,
+                "max_sequence_length": self.max_sequence_length,
+            }
+        )
+        return config
+
+    @property
+    def token_embedding(self):
+        return self.get_layer("token_embedding")
+
+    @classmethod
+    def create_layout_map(cls, mesh):
+
+        # We assert the mesh is 2D, and assume the first mesh dim is for data
+        # parallel and the second dim is for model parallel.
+        mesh_shape = mesh.shape()
+        if len(mesh_shape) != 2:
+            raise ValueError(
+                f"Expect to create layout based on 2D mesh, received {mesh}"
+            )
+        _, model_dim = mesh.dim_names
+        unshard_dim = dtensor.UNSHARDED
+
+        layout_map = keras.dtensor.experimental.LayoutMap(mesh=mesh)
+        # Embedding sharding
+        layout_map[r".*embeddings"] = Layout([unshard_dim, model_dim], mesh)
+
+        # Transformer block sharding
+        layout_map[r".*_(query|key|value)_dense.kernel"] = Layout(
+            [unshard_dim, unshard_dim, model_dim], mesh
+        )
+        layout_map[r".*_(query|key|value)_dense.bias"] = Layout(
+            [model_dim, unshard_dim], mesh
+        )
+        layout_map[r".*_feedforward_intermediate_dense.kernel"] = Layout(
+            [unshard_dim, model_dim], mesh
+        )
+        layout_map[r".*_feedforward_intermediate_dense.bias"] = Layout(
+            [model_dim], mesh
+        )
+        layout_map[r".*_feedforward_output_dense.kernel"] = Layout(
+            [model_dim, unshard_dim], mesh
+        )
+        layout_map[r".*_feedforward_output_dense.bias"] = Layout(
+            [unshard_dim], mesh
+        )
+        return layout_map