New TF embeddings (cleaner and faster) (#9418)
* Create new embeddings + add to BERT * Add Albert * Add DistilBert * Add Albert + Electra + Funnel * Add Longformer + Lxmert * Add last models * Apply style * Update the template * Remove unused imports * Rename attribute * Import embeddings in their own model file * Replace word_embeddings per weight * fix naming * Fix Albert * Fix Albert * Fix Longformer * Fix Lxmert Mobilebert and MPNet * Fix copy * Fix template * Update the get weights function * Update src/transformers/modeling_tf_utils.py Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * Update src/transformers/models/electra/modeling_tf_electra.py Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * address Sylvain's comments Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
This commit is contained in:
Julien Plu
@@ -65,6 +65,127 @@ TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST = [
|
||||
]
|
||||
|
||||
|
||||
# Copied from transformers.models.bert.modeling_tf_bert.TFBertWordEmbeddings
|
||||
class TFRobertaWordEmbeddings(tf.keras.layers.Layer):
|
||||
def __init__(self, vocab_size: int, hidden_size: int, initializer_range: float, **kwargs):
|
||||
super().__init__(**kwargs)
|
||||
|
||||
self.vocab_size = vocab_size
|
||||
self.hidden_size = hidden_size
|
||||
self.initializer_range = initializer_range
|
||||
|
||||
def build(self, input_shape):
|
||||
self.weight = self.add_weight(
|
||||
name="weight",
|
||||
shape=[self.vocab_size, self.hidden_size],
|
||||
initializer=get_initializer(initializer_range=self.initializer_range),
|
||||
)
|
||||
|
||||
super().build(input_shape=input_shape)
|
||||
|
||||
def get_config(self):
|
||||
config = {
|
||||
"vocab_size": self.vocab_size,
|
||||
"hidden_size": self.hidden_size,
|
||||
"initializer_range": self.initializer_range,
|
||||
}
|
||||
base_config = super().get_config()
|
||||
|
||||
return dict(list(base_config.items()) + list(config.items()))
|
||||
|
||||
def call(self, input_ids):
|
||||
flat_input_ids = tf.reshape(tensor=input_ids, shape=[-1])
|
||||
embeddings = tf.gather(params=self.weight, indices=flat_input_ids)
|
||||
embeddings = tf.reshape(
|
||||
tensor=embeddings, shape=tf.concat(values=[shape_list(tensor=input_ids), [self.hidden_size]], axis=0)
|
||||
)
|
||||
|
||||
embeddings.set_shape(shape=input_ids.shape.as_list() + [self.hidden_size])
|
||||
|
||||
return embeddings
|
||||
|
||||
|
||||
# Copied from transformers.models.bert.modeling_tf_bert.TFBertTokenTypeEmbeddings
|
||||
class TFRobertaTokenTypeEmbeddings(tf.keras.layers.Layer):
|
||||
def __init__(self, type_vocab_size: int, hidden_size: int, initializer_range: float, **kwargs):
|
||||
super().__init__(**kwargs)
|
||||
|
||||
self.type_vocab_size = type_vocab_size
|
||||
self.hidden_size = hidden_size
|
||||
self.initializer_range = initializer_range
|
||||
|
||||
def build(self, input_shape):
|
||||
self.token_type_embeddings = self.add_weight(
|
||||
name="embeddings",
|
||||
shape=[self.type_vocab_size, self.hidden_size],
|
||||
initializer=get_initializer(initializer_range=self.initializer_range),
|
||||
)
|
||||
|
||||
super().build(input_shape=input_shape)
|
||||
|
||||
def get_config(self):
|
||||
config = {
|
||||
"type_vocab_size": self.type_vocab_size,
|
||||
"hidden_size": self.hidden_size,
|
||||
"initializer_range": self.initializer_range,
|
||||
}
|
||||
base_config = super().get_config()
|
||||
|
||||
return dict(list(base_config.items()) + list(config.items()))
|
||||
|
||||
def call(self, token_type_ids):
|
||||
flat_token_type_ids = tf.reshape(tensor=token_type_ids, shape=[-1])
|
||||
one_hot_data = tf.one_hot(indices=flat_token_type_ids, depth=self.type_vocab_size, dtype=self._compute_dtype)
|
||||
embeddings = tf.matmul(a=one_hot_data, b=self.token_type_embeddings)
|
||||
embeddings = tf.reshape(
|
||||
tensor=embeddings, shape=tf.concat(values=[shape_list(tensor=token_type_ids), [self.hidden_size]], axis=0)
|
||||
)
|
||||
|
||||
embeddings.set_shape(shape=token_type_ids.shape.as_list() + [self.hidden_size])
|
||||
|
||||
return embeddings
|
||||
|
||||
|
||||
# Copied from transformers.models.longformer.modeling_tf_longformer.TFLongformerPositionEmbeddings
|
||||
class TFRobertaPositionEmbeddings(tf.keras.layers.Layer):
|
||||
def __init__(self, max_position_embeddings: int, hidden_size: int, initializer_range: float, **kwargs):
|
||||
super().__init__(**kwargs)
|
||||
|
||||
self.max_position_embeddings = max_position_embeddings
|
||||
self.hidden_size = hidden_size
|
||||
self.initializer_range = initializer_range
|
||||
|
||||
def build(self, input_shape):
|
||||
self.position_embeddings = self.add_weight(
|
||||
name="embeddings",
|
||||
shape=[self.max_position_embeddings, self.hidden_size],
|
||||
initializer=get_initializer(initializer_range=self.initializer_range),
|
||||
)
|
||||
|
||||
super().build(input_shape)
|
||||
|
||||
def get_config(self):
|
||||
config = {
|
||||
"max_position_embeddings": self.max_position_embeddings,
|
||||
"hidden_size": self.hidden_size,
|
||||
"initializer_range": self.initializer_range,
|
||||
}
|
||||
base_config = super().get_config()
|
||||
|
||||
return dict(list(base_config.items()) + list(config.items()))
|
||||
|
||||
def call(self, position_ids):
|
||||
flat_position_ids = tf.reshape(tensor=position_ids, shape=[-1])
|
||||
embeddings = tf.gather(params=self.position_embeddings, indices=flat_position_ids)
|
||||
embeddings = tf.reshape(
|
||||
tensor=embeddings, shape=tf.concat(values=[shape_list(tensor=position_ids), [self.hidden_size]], axis=0)
|
||||
)
|
||||
|
||||
embeddings.set_shape(shape=position_ids.shape.as_list() + [self.hidden_size])
|
||||
|
||||
return embeddings
|
||||
|
||||
|
||||
class TFRobertaEmbeddings(tf.keras.layers.Layer):
|
||||
"""
|
||||
Same as BertEmbeddings with a tiny tweak for positional embeddings indexing.
|
||||
@@ -74,52 +195,48 @@ class TFRobertaEmbeddings(tf.keras.layers.Layer):
|
||||
super().__init__(**kwargs)
|
||||
|
||||
self.padding_idx = 1
|
||||
self.vocab_size = config.vocab_size
|
||||
self.hidden_size = config.hidden_size
|
||||
self.initializer_range = config.initializer_range
|
||||
self.position_embeddings = tf.keras.layers.Embedding(
|
||||
config.max_position_embeddings,
|
||||
config.hidden_size,
|
||||
embeddings_initializer=get_initializer(self.initializer_range),
|
||||
self.word_embeddings = TFRobertaWordEmbeddings(
|
||||
vocab_size=config.vocab_size,
|
||||
hidden_size=config.hidden_size,
|
||||
initializer_range=config.initializer_range,
|
||||
name="word_embeddings",
|
||||
)
|
||||
self.position_embeddings = TFRobertaPositionEmbeddings(
|
||||
max_position_embeddings=config.max_position_embeddings,
|
||||
hidden_size=config.hidden_size,
|
||||
initializer_range=config.initializer_range,
|
||||
name="position_embeddings",
|
||||
)
|
||||
self.token_type_embeddings = tf.keras.layers.Embedding(
|
||||
config.type_vocab_size,
|
||||
config.hidden_size,
|
||||
embeddings_initializer=get_initializer(self.initializer_range),
|
||||
self.token_type_embeddings = TFRobertaTokenTypeEmbeddings(
|
||||
type_vocab_size=config.type_vocab_size,
|
||||
hidden_size=config.hidden_size,
|
||||
initializer_range=config.initializer_range,
|
||||
name="token_type_embeddings",
|
||||
)
|
||||
|
||||
# self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
|
||||
# any TensorFlow checkpoint file
|
||||
self.embeddings_sum = tf.keras.layers.Add()
|
||||
self.LayerNorm = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_eps, name="LayerNorm")
|
||||
self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob)
|
||||
self.dropout = tf.keras.layers.Dropout(rate=config.hidden_dropout_prob)
|
||||
|
||||
def build(self, input_shape):
|
||||
"""Build shared word embedding layer """
|
||||
with tf.name_scope("word_embeddings"):
|
||||
# Create and initialize weights. The random normal initializer was chosen
|
||||
# arbitrarily, and works well.
|
||||
self.word_embeddings = self.add_weight(
|
||||
"weight",
|
||||
shape=[self.vocab_size, self.hidden_size],
|
||||
initializer=get_initializer(self.initializer_range),
|
||||
)
|
||||
|
||||
super().build(input_shape)
|
||||
|
||||
def create_position_ids_from_input_ids(self, x):
|
||||
def create_position_ids_from_input_ids(self, input_ids):
|
||||
"""
|
||||
Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding
|
||||
symbols are ignored. This is modified from fairseq's `utils.make_positions`.
|
||||
|
||||
Args:
|
||||
x: tf.Tensor
|
||||
input_ids: tf.Tensor
|
||||
|
||||
Returns: tf.Tensor
|
||||
"""
|
||||
mask = tf.cast(tf.math.not_equal(x, self.padding_idx), dtype=tf.int32)
|
||||
incremental_indices = tf.math.cumsum(mask, axis=1) * mask
|
||||
input_ids_shape = shape_list(tensor=input_ids)
|
||||
|
||||
# multiple choice has 3 dimensions
|
||||
if len(input_ids_shape) == 3:
|
||||
input_ids = tf.reshape(
|
||||
tensor=input_ids, shape=(input_ids_shape[0] * input_ids_shape[1], input_ids_shape[2])
|
||||
)
|
||||
|
||||
mask = tf.cast(x=tf.math.not_equal(x=input_ids, y=self.padding_idx), dtype=input_ids.dtype)
|
||||
incremental_indices = tf.math.cumsum(x=mask, axis=1) * mask
|
||||
|
||||
return incremental_indices + self.padding_idx
|
||||
|
||||
@@ -132,96 +249,41 @@ class TFRobertaEmbeddings(tf.keras.layers.Layer):
|
||||
|
||||
Returns: tf.Tensor
|
||||
"""
|
||||
seq_length = shape_list(inputs_embeds)[1]
|
||||
position_ids = tf.range(self.padding_idx + 1, seq_length + self.padding_idx + 1, dtype=tf.int32)[tf.newaxis, :]
|
||||
batch_size, seq_length = shape_list(tensor=inputs_embeds)[:2]
|
||||
position_ids = tf.range(start=self.padding_idx + 1, limit=seq_length + self.padding_idx + 1)[tf.newaxis, :]
|
||||
|
||||
return position_ids
|
||||
return tf.tile(input=position_ids, multiples=(batch_size, 1))
|
||||
|
||||
def call(
|
||||
self,
|
||||
input_ids=None,
|
||||
position_ids=None,
|
||||
token_type_ids=None,
|
||||
inputs_embeds=None,
|
||||
mode="embedding",
|
||||
training=False,
|
||||
):
|
||||
def call(self, input_ids=None, position_ids=None, token_type_ids=None, inputs_embeds=None, training=False):
|
||||
"""
|
||||
Get token embeddings of inputs.
|
||||
|
||||
Args:
|
||||
inputs: list of three int64 tensors with shape [batch_size, length]: (input_ids, position_ids, token_type_ids)
|
||||
mode: string, a valid value is one of "embedding" and "linear".
|
||||
Applies embedding based on inputs tensor.
|
||||
|
||||
Returns:
|
||||
outputs: If mode == "embedding", output embedding tensor, float32 with shape [batch_size, length,
|
||||
embedding_size]; if mode == "linear", output linear tensor, float32 with shape [batch_size, length,
|
||||
vocab_size].
|
||||
|
||||
Raises:
|
||||
ValueError: if mode is not valid.
|
||||
|
||||
Shared weights logic adapted from
|
||||
https://github.com/tensorflow/models/blob/a009f4fb9d2fc4949e32192a944688925ef78659/official/transformer/v2/embedding_layer.py#L24
|
||||
final_embeddings (:obj:`tf.Tensor`): output embedding tensor.
|
||||
"""
|
||||
if mode == "embedding":
|
||||
return self._embedding(input_ids, position_ids, token_type_ids, inputs_embeds, training=training)
|
||||
elif mode == "linear":
|
||||
return self._linear(input_ids)
|
||||
else:
|
||||
raise ValueError("mode {} is not valid.".format(mode))
|
||||
|
||||
def _embedding(self, input_ids, position_ids, token_type_ids, inputs_embeds, training=False):
|
||||
"""Applies embedding based on inputs tensor."""
|
||||
assert not (input_ids is None and inputs_embeds is None)
|
||||
|
||||
if input_ids is not None:
|
||||
inputs_embeds = self.word_embeddings(input_ids=input_ids)
|
||||
|
||||
if token_type_ids is None:
|
||||
input_shape = shape_list(tensor=inputs_embeds)[:-1]
|
||||
token_type_ids = tf.fill(dims=input_shape, value=0)
|
||||
|
||||
if position_ids is None:
|
||||
if input_ids is not None:
|
||||
# Create the position ids from the input token ids. Any padded tokens remain padded.
|
||||
position_ids = self.create_position_ids_from_input_ids(input_ids)
|
||||
position_ids = self.create_position_ids_from_input_ids(input_ids=input_ids)
|
||||
else:
|
||||
position_ids = self.create_position_ids_from_inputs_embeds(inputs_embeds)
|
||||
position_ids = self.create_position_ids_from_inputs_embeds(inputs_embeds=inputs_embeds)
|
||||
|
||||
if input_ids is not None:
|
||||
input_shape = shape_list(input_ids)
|
||||
else:
|
||||
input_shape = shape_list(inputs_embeds)[:-1]
|
||||
position_embeds = self.position_embeddings(position_ids=position_ids)
|
||||
token_type_embeds = self.token_type_embeddings(token_type_ids=token_type_ids)
|
||||
final_embeddings = self.embeddings_sum(inputs=[inputs_embeds, position_embeds, token_type_embeds])
|
||||
final_embeddings = self.LayerNorm(inputs=final_embeddings)
|
||||
final_embeddings = self.dropout(inputs=final_embeddings, training=training)
|
||||
|
||||
seq_length = input_shape[1]
|
||||
|
||||
if position_ids is None:
|
||||
position_ids = tf.range(seq_length, dtype=tf.int32)[tf.newaxis, :]
|
||||
|
||||
if token_type_ids is None:
|
||||
token_type_ids = tf.fill(input_shape, 0)
|
||||
|
||||
if inputs_embeds is None:
|
||||
inputs_embeds = tf.gather(self.word_embeddings, input_ids)
|
||||
|
||||
position_embeddings = tf.cast(self.position_embeddings(position_ids), inputs_embeds.dtype)
|
||||
token_type_embeddings = tf.cast(self.token_type_embeddings(token_type_ids), inputs_embeds.dtype)
|
||||
embeddings = inputs_embeds + position_embeddings + token_type_embeddings
|
||||
embeddings = self.LayerNorm(embeddings)
|
||||
embeddings = self.dropout(embeddings, training=training)
|
||||
|
||||
return embeddings
|
||||
|
||||
def _linear(self, inputs):
|
||||
"""
|
||||
Computes logits by running inputs through a linear layer.
|
||||
|
||||
Args:
|
||||
inputs: A float32 tensor with shape [batch_size, length, hidden_size]
|
||||
|
||||
Returns:
|
||||
float32 tensor with shape [batch_size, length, vocab_size].
|
||||
"""
|
||||
batch_size = shape_list(inputs)[0]
|
||||
length = shape_list(inputs)[1]
|
||||
x = tf.reshape(inputs, [-1, self.hidden_size])
|
||||
logits = tf.matmul(x, self.word_embeddings, transpose_b=True)
|
||||
|
||||
return tf.reshape(logits, [batch_size, length, self.vocab_size])
|
||||
return final_embeddings
|
||||
|
||||
|
||||
# Copied from transformers.models.bert.modeling_tf_bert.TFBertPooler
|
||||
@@ -245,7 +307,7 @@ class TFRobertaPooler(tf.keras.layers.Layer):
|
||||
return pooled_output
|
||||
|
||||
|
||||
# Copied from transformers.models.bert.modeling_tf_bert.TFBertSelfAttention with Bert->Roberta
|
||||
# Copied from transformers.models.bert.modeling_tf_bert.TFBertSelfAttention
|
||||
class TFRobertaSelfAttention(tf.keras.layers.Layer):
|
||||
def __init__(self, config, **kwargs):
|
||||
super().__init__(**kwargs)
|
||||
@@ -256,8 +318,8 @@ class TFRobertaSelfAttention(tf.keras.layers.Layer):
|
||||
f"of attention heads ({config.num_attention_heads})"
|
||||
)
|
||||
|
||||
self.num_attention_heads = config.num_attention_heads
|
||||
self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
|
||||
|
||||
self.query = tf.keras.layers.experimental.EinsumDense(
|
||||
equation="abc,cde->abde",
|
||||
output_shape=(None, config.num_attention_heads, self.attention_head_size),
|
||||
@@ -293,7 +355,7 @@ class TFRobertaSelfAttention(tf.keras.layers.Layer):
|
||||
attention_scores = tf.einsum("aecd,abcd->acbe", key_layer, query_layer)
|
||||
|
||||
if attention_mask is not None:
|
||||
# Apply the attention mask is (precomputed for all layers in TFRobertaModel call() function)
|
||||
# Apply the attention mask is (precomputed for all layers in TFBertModel call() function)
|
||||
attention_scores = attention_scores + attention_mask
|
||||
|
||||
# Normalize the attention scores to probabilities.
|
||||
@@ -324,9 +386,9 @@ class TFRobertaSelfOutput(tf.keras.layers.Layer):
|
||||
f"of attention heads ({config.num_attention_heads})"
|
||||
)
|
||||
|
||||
self.num_attention_heads = config.num_attention_heads
|
||||
self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
|
||||
self.all_head_size = self.num_attention_heads * self.attention_head_size
|
||||
self.all_head_size = config.num_attention_heads * self.attention_head_size
|
||||
|
||||
self.dense = tf.keras.layers.experimental.EinsumDense(
|
||||
equation="abcd,cde->abe",
|
||||
output_shape=(None, self.all_head_size),
|
||||
@@ -499,12 +561,12 @@ class TFRobertaMainLayer(tf.keras.layers.Layer):
|
||||
|
||||
# Copied from transformers.models.bert.modeling_tf_bert.TFBertMainLayer.get_input_embeddings
|
||||
def get_input_embeddings(self):
|
||||
return self.embeddings
|
||||
return self.embeddings.word_embeddings
|
||||
|
||||
# Copied from transformers.models.bert.modeling_tf_bert.TFBertMainLayer.set_input_embeddings
|
||||
def set_input_embeddings(self, value):
|
||||
self.embeddings.word_embeddings = value
|
||||
self.embeddings.vocab_size = shape_list(value)[0]
|
||||
self.embeddings.word_embeddings.weight = value
|
||||
self.embeddings.word_embeddings.vocab_size = shape_list(value)[0]
|
||||
|
||||
# Copied from transformers.models.bert.modeling_tf_bert.TFBertMainLayer._prune_heads
|
||||
def _prune_heads(self, heads_to_prune):
|
||||
@@ -814,6 +876,7 @@ class TFRobertaLMHead(tf.keras.layers.Layer):
|
||||
super().__init__(**kwargs)
|
||||
|
||||
self.vocab_size = config.vocab_size
|
||||
self.hidden_size = config.hidden_size
|
||||
self.dense = tf.keras.layers.Dense(
|
||||
config.hidden_size, kernel_initializer=get_initializer(config.initializer_range), name="dense"
|
||||
)
|
||||
@@ -833,7 +896,7 @@ class TFRobertaLMHead(tf.keras.layers.Layer):
|
||||
return self.decoder
|
||||
|
||||
def set_output_embeddings(self, value):
|
||||
self.decoder.word_embeddings = value
|
||||
self.decoder.weight = value
|
||||
self.decoder.vocab_size = shape_list(value)[0]
|
||||
|
||||
def get_bias(self):
|
||||
@@ -849,7 +912,11 @@ class TFRobertaLMHead(tf.keras.layers.Layer):
|
||||
hidden_states = self.layer_norm(hidden_states)
|
||||
|
||||
# project back to size of vocabulary with bias
|
||||
hidden_states = self.decoder(hidden_states, mode="linear") + self.bias
|
||||
seq_length = shape_list(tensor=hidden_states)[1]
|
||||
hidden_states = tf.reshape(tensor=hidden_states, shape=[-1, self.hidden_size])
|
||||
hidden_states = tf.matmul(a=hidden_states, b=self.decoder.weight, transpose_b=True)
|
||||
hidden_states = tf.reshape(tensor=hidden_states, shape=[-1, seq_length, self.vocab_size])
|
||||
hidden_states = tf.nn.bias_add(value=hidden_states, bias=self.bias)
|
||||
|
||||
return hidden_states
|
||||
|
||||
@@ -863,7 +930,7 @@ class TFRobertaForMaskedLM(TFRobertaPreTrainedModel, TFMaskedLanguageModelingLos
|
||||
super().__init__(config, *inputs, **kwargs)
|
||||
|
||||
self.roberta = TFRobertaMainLayer(config, add_pooling_layer=False, name="roberta")
|
||||
self.lm_head = TFRobertaLMHead(config, self.roberta.embeddings, name="lm_head")
|
||||
self.lm_head = TFRobertaLMHead(config, self.roberta.embeddings.word_embeddings, name="lm_head")
|
||||
|
||||
def get_lm_head(self):
|
||||
return self.lm_head
|
||||
|
||||
Reference in New Issue
Block a user