Adding GPTNeoXBackbone

keras_nlp/models/__init__.py

@@ -78,6 +78,7 @@
 )
 from keras_nlp.models.gpt2.gpt2_preprocessor import GPT2Preprocessor
 from keras_nlp.models.gpt2.gpt2_tokenizer import GPT2Tokenizer
+from keras_nlp.models.gpt_neo_x.gpt_neo_x_backbone import GPTNeoXBackbone
 from keras_nlp.models.opt.opt_backbone import OPTBackbone
 from keras_nlp.models.opt.opt_causal_lm import OPTCausalLM
 from keras_nlp.models.opt.opt_causal_lm_preprocessor import (

keras_nlp/models/gpt_neo_x/__init__.py

@@ -0,0 +1,13 @@
+# 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.

keras_nlp/models/gpt_neo_x/gpt_neo_x_attention.py

@@ -0,0 +1,175 @@
+# 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.models.gpt_neo_x.rotary_embedding import RotaryEmbedding
+from keras_nlp.utils.keras_utils import clone_initializer
+
+
+class GPTNeoXAttention(keras.layers.Layer):
+    """GPTNeoXAttention layer.
+
+    This is an implementation of disentangled attention as described in the
+    paper ["GPT-NeoX-20B: An Open-Source Autoregressive Language Model"](https://arxiv.org/abs/2204.06745).
+    Effectively, this layer implements Multi-Head Self Attention with rotary embedding,
+
+    Args:
+        num_heads: int. Number of attention heads.
+        hidden_dim: int. Hidden dimension of the input, i.e., `hidden_states`.
+        max_position_embeddings: int, defaults to 512. The maximum input
+            sequence length.
+        bucket_size: int, defaults to 256. The size of the relative position
+            buckets. Generally equal to `max_sequence_length // 2`.
+        dropout: float, defaults to 0.1. Dropout probability.
+        kernel_initializer: string or `keras.initializers` initializer,
+            defaults to "glorot_uniform". The kernel initializer for
+            the dense layers.
+        bias_initializer: string or `keras.initializers` initializer,
+            defaults to "zeros". The bias initializer for the dense layers.
+        rotary_max_wavelength: int. The maximum angular wavelength of the sine/cosine
+            curves, for rotary embeddings. Defaults to 10000.
+        rotary_percentage: float. The percentage by which query, key, value matrices are
+            to be rotated
+    """
+
+    def __init__(
+        self,
+        num_heads,
+        hidden_dim,
+        dropout=0.1,
+        max_sequence_length=512,
+        kernel_initializer="glorot_uniform",
+        bias_initializer="zeros",
+        rotary_percentage=0.25,
+        rotary_max_wavelength=10000,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.num_heads = num_heads
+        self.hidden_dim = hidden_dim
+        self.rotary_percentage = rotary_percentage
+        self.dropout = dropout
+        self.attn_head_size = hidden_dim // num_heads
+        self.rotary_ndims = int(self.attn_head_size * rotary_percentage)
+        self.rotary_max_wavelength = rotary_max_wavelength
+        self.max_sequence_length = max_sequence_length
+        self.rotary_embedding = RotaryEmbedding(
+            self.rotary_ndims, rotary_max_wavelength
+        )
+        self._kernel_initializer = keras.initializers.get(kernel_initializer)
+        self._bias_initializer = keras.initializers.get(bias_initializer)
+
+        self._qkv_dense = keras.layers.EinsumDense(
+            equation="abc,cde->abde",
+            output_shape=(None, self.num_heads, 3 * self.attn_head_size),
+            bias_axes="de",
+            **self._get_common_kwargs_for_sublayer(use_bias=True),
+            name="query",
+        )
+
+        self._attn_dropout_layer = keras.layers.Dropout(
+            self.dropout, name="attention_dropout"
+        )
+
+        self._softmax = keras.layers.Softmax(axis=-1, name="attention_softmax")
+
+        # Output.
+        self._output_dense = keras.layers.EinsumDense(
+            equation="abc,cd->abd",
+            output_shape=(None, self.hidden_dim),
+            bias_axes="d",
+            **self._get_common_kwargs_for_sublayer(use_bias=True),
+            name="attention_output",
+        )
+
+    def _get_common_kwargs_for_sublayer(self, use_bias=True):
+        common_kwargs = {}
+
+        kernel_initializer = clone_initializer(self._kernel_initializer)
+        bias_initializer = clone_initializer(self._bias_initializer)
+
+        common_kwargs["kernel_initializer"] = kernel_initializer
+        if use_bias:
+            common_kwargs["bias_initializer"] = bias_initializer
+
+        return common_kwargs
+
+    def _masked_softmax(self, attention_scores, attention_mask=None):
+        if attention_mask is not None:
+            mask_expansion_axis = -3
+            for _ in range(
+                attention_scores.shape.rank - attention_mask.shape.rank
+            ):
+                attention_mask = tf.expand_dims(
+                    attention_mask, axis=mask_expansion_axis
+                )
+        return self._softmax(attention_scores, attention_mask)
+
+    def _compute_attention(
+        self, query, key, value, attention_mask=None, training=None
+    ):
+        attention_scores = tf.einsum("aecd,abcd->acbe", key, query)
+        attention_scores /= tf.sqrt(self.attn_head_size)
+
+        attention_scores = self._masked_softmax(
+            attention_scores, attention_mask
+        )
+        attention_scores = self._attn_dropout_layer(
+            attention_scores, training=training
+        )
+        attention_output = tf.einsum("acbe,aecd->abcd", attention_scores, value)
+
+        return attention_output, attention_scores
+
+    def call(
+        self,
+        hidden_states,
+        attention_mask,
+        return_attention_scores=False,
+        training=None,
+    ):
+        query_key_value = self._qkv_dense(hidden_states)
+
+        query = query_key_value[..., : self.attn_head_size]
+        key = query_key_value[
+            ..., self.attn_head_size : 2 * self.attn_head_size
+        ]
+        value = query_key_value[..., 2 * self.attn_head_size :]
+
+        query, key = self.rotary_embedding(query, key)
+
+        attention_output, attention_scores = self._compute_attention(
+            query=query,
+            key=key,
+            value=value,
+            attention_mask=attention_mask,
+            training=training,
+        )
+
+        # Reshape `attention_output` to `(batch_size, sequence_length, hidden_dim)`.
+        attention_output = tf.reshape(
+            attention_output,
+            [
+                tf.shape(attention_output)[0],
+                tf.shape(attention_output)[1],
+                self.hidden_dim,
+            ],
+        )
+
+        attention_output = self._output_dense(attention_output)
+
+        if return_attention_scores:
+            return attention_output, attention_scores
+        return attention_output

keras_nlp/models/gpt_neo_x/gpt_neo_x_backbone.py

@@ -0,0 +1,160 @@
+# 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.api_export import keras_nlp_export
+from keras_nlp.models.backbone import Backbone
+from keras_nlp.models.gpt_neo_x.gpt_neo_x_decoder import GPTNeoXDecoder
+
+
+def _gpt_neo_x_kernel_initializer(stddev=0.02):
+    return keras.initializers.RandomNormal(stddev=stddev)
+
+
+@keras_nlp_export("keras_nlp.models.GPTNeoXBackbone")
+class GPTNeoXBackbone(Backbone):
+    """GPT-2 core network with hyperparameters.
+
+    This network implements a Transformer-based decoder network,
+    Generative Pretrained Transformer-Neo-X (GPTNeoX), as described in
+    ["GPT-NeoX-20B: An Open-Source Autoregressive Language Model"](https://arxiv.org/abs/2204.06745).
+    It includes the embedding lookups and transformer layers.
+
+    The default constructor gives a fully customizable, randomly initialized
+    GPT-NeoX model with any number of layers, heads, and embedding
+    dimensions.
+
+    Disclaimer: Pre-trained models are provided on an "as is" basis, without
+    warranties or conditions of any kind. The underlying model is provided by a
+    third party and subject to a separate license, available
+    [here](https://github.com/EleutherAI/pythiarota).
+
+    Args:
+        vocabulary_size: int. The size of the token vocabulary.
+        num_layers: int. The number of transformer layers.
+        num_heads: int. The number of attention heads for each transformer.
+            The hidden size must be divisible by the number of attention heads.
+        hidden_dim: int. The size of the transformer encoding and pooler layers.
+        intermediate_dim: int. The output dimension of the first Dense layer in
+            a two-layer feedforward network for each transformer.
+        dropout: float. Dropout probability for the Transformer encoder.
+        layer_norm_epsilon: float. a value added to the denominator for numerical stability.
+           Default: 1e-5
+        rotary_max_wavelength: int. The maximum angular wavelength of the sine/cosine
+           curves, for rotary embeddings. Defaults to 10000.
+        rotary_percentage: float. The percentage by which query, key, value matrices are
+           to be rotated
+        max_sequence_length: int. The maximum sequence length that this encoder
+            can consume. If `None`, `max_sequence_length` uses the value from
+            sequence length. This determines the variable shape for positional
+            embeddings.
+        **kwargs: other keyword arguments.
+    """
+
+    def __init__(
+        self,
+        vocabulary_size,
+        num_layers,
+        num_heads,
+        hidden_dim,
+        intermediate_dim,
+        dropout=0.0,
+        rotary_percentage=0.25,
+        rotary_max_wavelength=10000,
+        layer_norm_epsilon=1e-5,
+        max_sequence_length=512,
+        **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_neo_x_kernel_initializer(stddev=0.01),
+            name="token_embedding",
+        )(token_ids)
+
+        x = keras.layers.Dropout(
+            dropout,
+            name="embeddings_dropout",
+        )(token_embedding)
+
+        # Apply successive transformer decoder blocks.
+        for i in range(num_layers):
+            x = GPTNeoXDecoder(
+                intermediate_dim=intermediate_dim,
+                num_heads=num_heads,
+                dropout=dropout,
+                max_sequence_length=max_sequence_length,
+                rotary_percentage=rotary_percentage,
+                rotary_max_wavelength=rotary_max_wavelength,
+                layer_norm_epsilon=layer_norm_epsilon,
+                activation=lambda x: keras.activations.gelu(
+                    x, approximate=True
+                ),
+                kernel_initializer=_gpt_neo_x_kernel_initializer(stddev=0.02),
+                name=f"transformer_layer_{i}",
+            )(x, decoder_padding_mask=padding_mask)
+
+        sequence_output = keras.layers.LayerNormalization(
+            name="layer_norm",
+            axis=-1,
+            epsilon=layer_norm_epsilon,
+            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
+        self.layer_norm_epsilon = layer_norm_epsilon
+
+    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,
+                "layer_norm_epsilon": self.layer_norm_epsilon,
+            }
+        )
+        return config
+
+    @property
+    def token_embedding(self):
+        return self.get_layer("token_embedding")