Merge remote-tracking branch 'upstream/master' into newavarecords

official/modeling/tf_utils.py

@@ -88,7 +88,6 @@ def is_special_none_tensor(tensor):
   return tensor.shape.ndims == 0 and tensor.dtype == tf.int32
 
 
-# TODO(hongkuny): consider moving custom string-map lookup to keras api.
 def get_activation(identifier):
   """Maps a identifier to a Python function, e.g., "relu" => `tf.nn.relu`.
 

official/nlp/albert/run_classifier.py

@@ -14,23 +14,61 @@
 # ==============================================================================
 """ALBERT classification finetuning runner in tf2.x."""
 
+
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
 import json
-
+import os
 from absl import app
 from absl import flags
+from absl import logging
 import tensorflow as tf
 
 from official.nlp.albert import configs as albert_configs
+from official.nlp.bert import bert_models
 from official.nlp.bert import run_classifier as run_classifier_bert
 from official.utils.misc import distribution_utils
 
+
 FLAGS = flags.FLAGS
 
 
+def predict(strategy, albert_config, input_meta_data, predict_input_fn):
+  """Function outputs both the ground truth predictions as .tsv files."""
+  with strategy.scope():
+    classifier_model = bert_models.classifier_model(
+        albert_config, input_meta_data['num_labels'])[0]
+    checkpoint = tf.train.Checkpoint(model=classifier_model)
+    latest_checkpoint_file = (
+        FLAGS.predict_checkpoint_path or
+        tf.train.latest_checkpoint(FLAGS.model_dir))
+    assert latest_checkpoint_file
+    logging.info('Checkpoint file %s found and restoring from '
+                 'checkpoint', latest_checkpoint_file)
+    checkpoint.restore(
+        latest_checkpoint_file).assert_existing_objects_matched()
+    preds, ground_truth = run_classifier_bert.get_predictions_and_labels(
+        strategy, classifier_model, predict_input_fn, return_probs=True)
+    output_predict_file = os.path.join(FLAGS.model_dir, 'test_results.tsv')
+    with tf.io.gfile.GFile(output_predict_file, 'w') as writer:
+      logging.info('***** Predict results *****')
+      for probabilities in preds:
+        output_line = '\t'.join(
+            str(class_probability)
+            for class_probability in probabilities) + '\n'
+        writer.write(output_line)
+    ground_truth_labels_file = os.path.join(FLAGS.model_dir,
+                                            'output_labels.tsv')
+    with tf.io.gfile.GFile(ground_truth_labels_file, 'w') as writer:
+      logging.info('***** Ground truth results *****')
+      for label in ground_truth:
+        output_line = '\t'.join(str(label)) + '\n'
+        writer.write(output_line)
+  return
+
+
 def main(_):
   with tf.io.gfile.GFile(FLAGS.input_meta_data_path, 'rb') as reader:
     input_meta_data = json.loads(reader.read().decode('utf-8'))
@@ -56,9 +94,14 @@ def main(_):
 
   albert_config = albert_configs.AlbertConfig.from_json_file(
       FLAGS.bert_config_file)
+  if FLAGS.mode == 'train_and_eval':
     run_classifier_bert.run_bert(strategy, input_meta_data, albert_config,
                                  train_input_fn, eval_input_fn)
-
+  elif FLAGS.mode == 'predict':
+    predict(strategy, albert_config, input_meta_data, eval_input_fn)
+  else:
+    raise ValueError('Unsupported mode is specified: %s' % FLAGS.mode)
+  return
 
 if __name__ == '__main__':
   flags.mark_flag_as_required('bert_config_file')

official/nlp/bert/export_tfhub.py

@@ -79,7 +79,7 @@ def export_bert_tfhub(bert_config: configs.BertConfig,
                  do_lower_case, vocab_file)
   core_model, encoder = create_bert_model(bert_config)
   checkpoint = tf.train.Checkpoint(model=encoder)
-  checkpoint.restore(model_checkpoint_path).assert_consumed()
+  checkpoint.restore(model_checkpoint_path).assert_existing_objects_matched()
   core_model.vocab_file = tf.saved_model.Asset(vocab_file)
   core_model.do_lower_case = tf.Variable(do_lower_case, trainable=False)
   core_model.save(hub_destination, include_optimizer=False, save_format="tf")

official/nlp/bert/model_saving_utils.py

@@ -55,14 +55,10 @@ def export_bert_model(model_export_path: typing.Text,
     raise ValueError('model must be a tf.keras.Model object.')
 
   if checkpoint_dir:
-    # Keras compile/fit() was used to save checkpoint using
-    # model.save_weights().
     if restore_model_using_load_weights:
       model_weight_path = os.path.join(checkpoint_dir, 'checkpoint')
       assert tf.io.gfile.exists(model_weight_path)
       model.load_weights(model_weight_path)
-
-    # tf.train.Checkpoint API was used via custom training loop logic.
     else:
       checkpoint = tf.train.Checkpoint(model=model)
 

official/nlp/bert/model_training_utils.py

@@ -99,7 +99,9 @@ def write_txt_summary(training_summary, summary_dir):
 
 
 @deprecation.deprecated(
-    None, 'This function is deprecated. Please use Keras compile/fit instead.')
+    None, 'This function is deprecated and we do not expect adding new '
+    'functionalities. Please do not have your code depending '
+    'on this library.')
 def run_customized_training_loop(
     # pylint: disable=invalid-name
     _sentinel=None,
@@ -557,7 +559,6 @@ def run_customized_training_loop(
     for metric in model.metrics:
       training_summary[metric.name] = _float_metric_value(metric)
     if eval_metrics:
-      # TODO(hongkuny): Cleans up summary reporting in text.
       training_summary['last_train_metrics'] = _float_metric_value(
           train_metrics[0])
       training_summary['eval_metrics'] = _float_metric_value(eval_metrics[0])

official/nlp/bert/run_classifier.py

@@ -343,7 +343,10 @@ def export_classifier(model_export_path, input_meta_data, bert_config,
   # Export uses float32 for now, even if training uses mixed precision.
   tf.keras.mixed_precision.experimental.set_policy('float32')
   classifier_model = bert_models.classifier_model(
-      bert_config, input_meta_data.get('num_labels', 1))[0]
+      bert_config,
+      input_meta_data.get('num_labels', 1),
+      hub_module_url=FLAGS.hub_module_url,
+      hub_module_trainable=False)[0]
 
   model_saving_utils.export_bert_model(
       model_export_path, model=classifier_model, checkpoint_dir=model_dir)

official/nlp/configs/bert.py

@@ -24,7 +24,6 @@ import tensorflow as tf
 
 from official.modeling import tf_utils
 from official.modeling.hyperparams import base_config
-from official.modeling.hyperparams import config_definitions as cfg
 from official.nlp.configs import encoders
 from official.nlp.modeling import layers
 from official.nlp.modeling.models import bert_pretrainer
@@ -43,7 +42,6 @@ class ClsHeadConfig(base_config.Config):
 @dataclasses.dataclass
 class BertPretrainerConfig(base_config.Config):
   """BERT encoder configuration."""
-  num_masked_tokens: int = 76
   encoder: encoders.TransformerEncoderConfig = (
       encoders.TransformerEncoderConfig())
   cls_heads: List[ClsHeadConfig] = dataclasses.field(default_factory=list)
@@ -56,45 +54,18 @@ def instantiate_classification_heads_from_cfgs(
     ] if cls_head_configs else []
 
 
-def instantiate_bertpretrainer_from_cfg(
+def instantiate_pretrainer_from_cfg(
     config: BertPretrainerConfig,
     encoder_network: Optional[tf.keras.Model] = None
-    ) -> bert_pretrainer.BertPretrainerV2:
+) -> bert_pretrainer.BertPretrainerV2:
   """Instantiates a BertPretrainer from the config."""
   encoder_cfg = config.encoder
   if encoder_network is None:
     encoder_network = encoders.instantiate_encoder_from_cfg(encoder_cfg)
   return bert_pretrainer.BertPretrainerV2(
-      config.num_masked_tokens,
       mlm_activation=tf_utils.get_activation(encoder_cfg.hidden_activation),
       mlm_initializer=tf.keras.initializers.TruncatedNormal(
           stddev=encoder_cfg.initializer_range),
       encoder_network=encoder_network,
       classification_heads=instantiate_classification_heads_from_cfgs(
           config.cls_heads))
-
-
-@dataclasses.dataclass
-class QADataConfig(cfg.DataConfig):
-  """Data config for question answering task (tasks/question_answering)."""
-  input_path: str = ""
-  global_batch_size: int = 48
-  is_training: bool = True
-  seq_length: int = 384
-
-
-@dataclasses.dataclass
-class QADevDataConfig(cfg.DataConfig):
-  """Dev Data config for queston answering (tasks/question_answering)."""
-  input_path: str = ""
-  input_preprocessed_data_path: str = ""
-  version_2_with_negative: bool = False
-  doc_stride: int = 128
-  global_batch_size: int = 48
-  is_training: bool = False
-  seq_length: int = 384
-  query_length: int = 64
-  drop_remainder: bool = False
-  vocab_file: str = ""
-  tokenization: str = "WordPiece"  # WordPiece or SentencePiece
-  do_lower_case: bool = True

official/nlp/configs/bert_test.py

@@ -26,7 +26,7 @@ class BertModelsTest(tf.test.TestCase):
   def test_network_invocation(self):
     config = bert.BertPretrainerConfig(
         encoder=encoders.TransformerEncoderConfig(vocab_size=10, num_layers=1))
-    _ = bert.instantiate_bertpretrainer_from_cfg(config)
+    _ = bert.instantiate_pretrainer_from_cfg(config)
 
     # Invokes with classification heads.
     config = bert.BertPretrainerConfig(
@@ -35,7 +35,7 @@ class BertModelsTest(tf.test.TestCase):
             bert.ClsHeadConfig(
                 inner_dim=10, num_classes=2, name="next_sentence")
         ])
-    _ = bert.instantiate_bertpretrainer_from_cfg(config)
+    _ = bert.instantiate_pretrainer_from_cfg(config)
 
     with self.assertRaises(ValueError):
       config = bert.BertPretrainerConfig(
@@ -47,7 +47,7 @@ class BertModelsTest(tf.test.TestCase):
               bert.ClsHeadConfig(
                   inner_dim=10, num_classes=2, name="next_sentence")
           ])
-      _ = bert.instantiate_bertpretrainer_from_cfg(config)
+      _ = bert.instantiate_pretrainer_from_cfg(config)
 
   def test_checkpoint_items(self):
     config = bert.BertPretrainerConfig(
@@ -56,9 +56,10 @@ class BertModelsTest(tf.test.TestCase):
             bert.ClsHeadConfig(
                 inner_dim=10, num_classes=2, name="next_sentence")
         ])
-    encoder = bert.instantiate_bertpretrainer_from_cfg(config)
-    self.assertSameElements(encoder.checkpoint_items.keys(),
-                            ["encoder", "next_sentence.pooler_dense"])
+    encoder = bert.instantiate_pretrainer_from_cfg(config)
+    self.assertSameElements(
+        encoder.checkpoint_items.keys(),
+        ["encoder", "masked_lm", "next_sentence.pooler_dense"])
 
 
 if __name__ == "__main__":

official/nlp/configs/electra.py

@@ -34,6 +34,8 @@ class ELECTRAPretrainerConfig(base_config.Config):
   sequence_length: int = 512
   num_classes: int = 2
   discriminator_loss_weight: float = 50.0
+  tie_embeddings: bool = True
+  disallow_correct: bool = False
   generator_encoder: encoders.TransformerEncoderConfig = (
       encoders.TransformerEncoderConfig())
   discriminator_encoder: encoders.TransformerEncoderConfig = (
@@ -60,23 +62,30 @@ def instantiate_pretrainer_from_cfg(
   """Instantiates ElectraPretrainer from the config."""
   generator_encoder_cfg = config.generator_encoder
   discriminator_encoder_cfg = config.discriminator_encoder
-  if generator_network is None:
-    generator_network = encoders.instantiate_encoder_from_cfg(
-        generator_encoder_cfg)
+  # Copy discriminator's embeddings to generator for easier model serialization.
   if discriminator_network is None:
     discriminator_network = encoders.instantiate_encoder_from_cfg(
         discriminator_encoder_cfg)
+  if generator_network is None:
+    if config.tie_embeddings:
+      embedding_layer = discriminator_network.get_embedding_layer()
+      generator_network = encoders.instantiate_encoder_from_cfg(
+          generator_encoder_cfg, embedding_layer=embedding_layer)
+    else:
+      generator_network = encoders.instantiate_encoder_from_cfg(
+          generator_encoder_cfg)
+
   return electra_pretrainer.ElectraPretrainer(
       generator_network=generator_network,
       discriminator_network=discriminator_network,
       vocab_size=config.generator_encoder.vocab_size,
       num_classes=config.num_classes,
       sequence_length=config.sequence_length,
-      last_hidden_dim=config.generator_encoder.hidden_size,
       num_token_predictions=config.num_masked_tokens,
       mlm_activation=tf_utils.get_activation(
           generator_encoder_cfg.hidden_activation),
       mlm_initializer=tf.keras.initializers.TruncatedNormal(
           stddev=generator_encoder_cfg.initializer_range),
       classification_heads=instantiate_classification_heads_from_cfgs(
-          config.cls_heads))
+          config.cls_heads),
+      disallow_correct=config.disallow_correct)

official/nlp/configs/encoders.py

@@ -17,12 +17,13 @@
 
 Includes configurations and instantiation methods.
 """
-
+from typing import Optional
 import dataclasses
 import tensorflow as tf
 
 from official.modeling import tf_utils
 from official.modeling.hyperparams import base_config
+from official.nlp.modeling import layers
 from official.nlp.modeling import networks
 
 
@@ -40,12 +41,47 @@ class TransformerEncoderConfig(base_config.Config):
   max_position_embeddings: int = 512
   type_vocab_size: int = 2
   initializer_range: float = 0.02
+  embedding_size: Optional[int] = None
 
 
 def instantiate_encoder_from_cfg(
-    config: TransformerEncoderConfig) -> networks.TransformerEncoder:
+    config: TransformerEncoderConfig,
+    encoder_cls=networks.TransformerEncoder,
+    embedding_layer: Optional[layers.OnDeviceEmbedding] = None):
   """Instantiate a Transformer encoder network from TransformerEncoderConfig."""
-  encoder_network = networks.TransformerEncoder(
+  if encoder_cls.__name__ == "EncoderScaffold":
+    embedding_cfg = dict(
+        vocab_size=config.vocab_size,
+        type_vocab_size=config.type_vocab_size,
+        hidden_size=config.hidden_size,
+        seq_length=None,
+        max_seq_length=config.max_position_embeddings,
+        initializer=tf.keras.initializers.TruncatedNormal(
+            stddev=config.initializer_range),
+        dropout_rate=config.dropout_rate,
+    )
+    hidden_cfg = dict(
+        num_attention_heads=config.num_attention_heads,
+        intermediate_size=config.intermediate_size,
+        intermediate_activation=tf_utils.get_activation(
+            config.hidden_activation),
+        dropout_rate=config.dropout_rate,
+        attention_dropout_rate=config.attention_dropout_rate,
+        kernel_initializer=tf.keras.initializers.TruncatedNormal(
+            stddev=config.initializer_range),
+    )
+    kwargs = dict(
+        embedding_cfg=embedding_cfg,
+        hidden_cfg=hidden_cfg,
+        num_hidden_instances=config.num_layers,
+        pooled_output_dim=config.hidden_size,
+        pooler_layer_initializer=tf.keras.initializers.TruncatedNormal(
+            stddev=config.initializer_range))
+    return encoder_cls(**kwargs)
+
+  if encoder_cls.__name__ != "TransformerEncoder":
+    raise ValueError("Unknown encoder network class. %s" % str(encoder_cls))
+  encoder_network = encoder_cls(
       vocab_size=config.vocab_size,
       hidden_size=config.hidden_size,
       num_layers=config.num_layers,
@@ -58,5 +94,7 @@ def instantiate_encoder_from_cfg(
       max_sequence_length=config.max_position_embeddings,
       type_vocab_size=config.type_vocab_size,
       initializer=tf.keras.initializers.TruncatedNormal(
-          stddev=config.initializer_range))
+          stddev=config.initializer_range),
+      embedding_width=config.embedding_size,
+      embedding_layer=embedding_layer)
   return encoder_network

official/nlp/data/create_finetuning_data.py

@@ -50,35 +50,41 @@ flags.DEFINE_string(
     "for the task.")
 
 flags.DEFINE_enum("classification_task_name", "MNLI",
-                  ["COLA", "MNLI", "MRPC", "QNLI", "QQP", "SST-2", "XNLI",
-                   "PAWS-X", "XTREME-XNLI", "XTREME-PAWS-X"],
+                  ["COLA", "MNLI", "MRPC", "PAWS-X", "QNLI", "QQP", "RTE",
+                   "SST-2", "STS-B", "WNLI", "XNLI", "XTREME-XNLI",
+                   "XTREME-PAWS-X"],
                   "The name of the task to train BERT classifier. The "
                   "difference between XTREME-XNLI and XNLI is: 1. the format "
                   "of input tsv files; 2. the dev set for XTREME is english "
                   "only and for XNLI is all languages combined. Same for "
                   "PAWS-X.")
 
-# XNLI task specific flag.
+# MNLI task-specific flag.
+flags.DEFINE_enum(
+    "mnli_type", "matched", ["matched", "mismatched"],
+    "The type of MNLI dataset.")
+
+# XNLI task-specific flag.
 flags.DEFINE_string(
     "xnli_language", "en",
-    "Language of training data for XNIL task. If the value is 'all', the data "
+    "Language of training data for XNLI task. If the value is 'all', the data "
     "of all languages will be used for training.")
 
-# PAWS-X task specific flag.
+# PAWS-X task-specific flag.
 flags.DEFINE_string(
     "pawsx_language", "en",
-    "Language of trainig data for PAWS-X task. If the value is 'all', the data "
+    "Language of training data for PAWS-X task. If the value is 'all', the data "
     "of all languages will be used for training.")
 
-# Retrieva task specific flags
+# Retrieval task-specific flags.
 flags.DEFINE_enum("retrieval_task_name", "bucc", ["bucc", "tatoeba"],
                   "The name of sentence retrieval task for scoring")
 
-# Tagging task specific flags
+# Tagging task-specific flags.
 flags.DEFINE_enum("tagging_task_name", "panx", ["panx", "udpos"],
                   "The name of BERT tagging (token classification) task.")
 
-# BERT Squad task specific flags.
+# BERT Squad task-specific flags.
 flags.DEFINE_string(
     "squad_data_file", None,
     "The input data file in for generating training data for BERT squad task.")
@@ -178,7 +184,8 @@ def generate_classifier_dataset():
         "cola":
             classifier_data_lib.ColaProcessor,
         "mnli":
-            classifier_data_lib.MnliProcessor,
+            functools.partial(classifier_data_lib.MnliProcessor,
+                              mnli_type=FLAGS.mnli_type),
         "mrpc":
             classifier_data_lib.MrpcProcessor,
         "qnli":
@@ -187,6 +194,8 @@ def generate_classifier_dataset():
         "rte": classifier_data_lib.RteProcessor,
         "sst-2":
             classifier_data_lib.SstProcessor,
+        "sts-b":
+            classifier_data_lib.StsBProcessor,
         "xnli":
             functools.partial(classifier_data_lib.XnliProcessor,
                               language=FLAGS.xnli_language),

official/nlp/data/create_pretraining_data.py

@@ -18,6 +18,7 @@ from __future__ import division
 from __future__ import print_function
 
 import collections
+import itertools
 import random
 
 from absl import app
@@ -48,6 +49,12 @@ flags.DEFINE_bool(
     "do_whole_word_mask", False,
     "Whether to use whole word masking rather than per-WordPiece masking.")
 
+flags.DEFINE_integer(
+    "max_ngram_size", None,
+    "Mask contiguous whole words (n-grams) of up to `max_ngram_size` using a "
+    "weighting scheme to favor shorter n-grams. "
+    "Note: `--do_whole_word_mask=True` must also be set when n-gram masking.")
+
 flags.DEFINE_bool(
     "gzip_compress", False,
     "Whether to use `GZIP` compress option to get compressed TFRecord files.")
@@ -192,7 +199,8 @@ def create_training_instances(input_files,
                               masked_lm_prob,
                               max_predictions_per_seq,
                               rng,
-                              do_whole_word_mask=False):
+                              do_whole_word_mask=False,
+                              max_ngram_size=None):
   """Create `TrainingInstance`s from raw text."""
   all_documents = [[]]
 
@@ -229,7 +237,7 @@ def create_training_instances(input_files,
           create_instances_from_document(
               all_documents, document_index, max_seq_length, short_seq_prob,
               masked_lm_prob, max_predictions_per_seq, vocab_words, rng,
-              do_whole_word_mask))
+              do_whole_word_mask, max_ngram_size))
 
   rng.shuffle(instances)
   return instances
@@ -238,7 +246,8 @@ def create_training_instances(input_files,
 def create_instances_from_document(
     all_documents, document_index, max_seq_length, short_seq_prob,
     masked_lm_prob, max_predictions_per_seq, vocab_words, rng,
-    do_whole_word_mask=False):
+    do_whole_word_mask=False,
+    max_ngram_size=None):
   """Creates `TrainingInstance`s for a single document."""
   document = all_documents[document_index]
 
@@ -337,7 +346,7 @@ def create_instances_from_document(
         (tokens, masked_lm_positions,
          masked_lm_labels) = create_masked_lm_predictions(
              tokens, masked_lm_prob, max_predictions_per_seq, vocab_words, rng,
-             do_whole_word_mask)
+             do_whole_word_mask, max_ngram_size)
         instance = TrainingInstance(
             tokens=tokens,
             segment_ids=segment_ids,
@@ -355,72 +364,238 @@ def create_instances_from_document(
 MaskedLmInstance = collections.namedtuple("MaskedLmInstance",
                                           ["index", "label"])
 
+# A _Gram is a [half-open) interval of token indices which form a word.
+# E.g.,
+#   words:  ["The", "doghouse"]
+#   tokens: ["The", "dog", "##house"]
+#   grams:  [(0,1), (1,3)]
+_Gram = collections.namedtuple("_Gram", ["begin", "end"])
+
+
+def _window(iterable, size):
+  """Helper to create a sliding window iterator with a given size.
+
+  E.g.,
+    input = [1, 2, 3, 4]
+    _window(input, 1) => [1], [2], [3], [4]
+    _window(input, 2) => [1, 2], [2, 3], [3, 4]
+    _window(input, 3) => [1, 2, 3], [2, 3, 4]
+    _window(input, 4) => [1, 2, 3, 4]
+    _window(input, 5) => None
+
+  Arguments:
+    iterable: elements to iterate over.
+    size: size of the window.
+
+  Yields:
+    Elements of `iterable` batched into a sliding window of length `size`.
+  """
+  i = iter(iterable)
+  window = []
+  try:
+    for e in range(0, size):
+      window.append(next(i))
+    yield window
+  except StopIteration:
+    # handle the case where iterable's length is less than the window size.
+    return
+  for e in i:
+    window = window[1:] + [e]
+    yield window
+
+
+def _contiguous(sorted_grams):
+  """Test whether a sequence of grams is contiguous.
+
+  Arguments:
+    sorted_grams: _Grams which are sorted in increasing order.
+  Returns:
+    True if `sorted_grams` are touching each other.
+
+  E.g.,
+    _contiguous([(1, 4), (4, 5), (5, 10)]) == True
+    _contiguous([(1, 2), (4, 5)]) == False
+  """
+  for a, b in _window(sorted_grams, 2):
+    if a.end != b.begin:
+      return False
+  return True
+
+
+def _masking_ngrams(grams, max_ngram_size, max_masked_tokens, rng):
+  """Create a list of masking {1, ..., n}-grams from a list of one-grams.
+
+  This is an extention of 'whole word masking' to mask multiple, contiguous
+  words such as (e.g., "the red boat").
+
+  Each input gram represents the token indices of a single word,
+     words:  ["the", "red", "boat"]
+     tokens: ["the", "red", "boa", "##t"]
+     grams:  [(0,1), (1,2), (2,4)]
+
+  For a `max_ngram_size` of three, possible outputs masks include:
+    1-grams: (0,1), (1,2), (2,4)
+    2-grams: (0,2), (1,4)
+    3-grams; (0,4)
+
+  Output masks will not overlap and contain less than `max_masked_tokens` total
+  tokens.  E.g., for the example above with `max_masked_tokens` as three,
+  valid outputs are,
+       [(0,1), (1,2)]  # "the", "red" covering two tokens
+       [(1,2), (2,4)]  # "red", "boa", "##t" covering three tokens
+
+  The length of the selected n-gram follows a zipf weighting to
+  favor shorter n-gram sizes (weight(1)=1, weight(2)=1/2, weight(3)=1/3, ...).
+
+  Arguments:
+    grams: List of one-grams.
+    max_ngram_size: Maximum number of contiguous one-grams combined to create
+      an n-gram.
+    max_masked_tokens: Maximum total number of tokens to be masked.
+    rng: `random.Random` generator.
+
+  Returns:
+    A list of n-grams to be used as masks.
+  """
+  if not grams:
+    return None
+
+  grams = sorted(grams)
+  num_tokens = grams[-1].end
+
+  # Ensure our grams are valid (i.e., they don't overlap).
+  for a, b in _window(grams, 2):
+    if a.end > b.begin:
+      raise ValueError("overlapping grams: {}".format(grams))
+
+  # Build map from n-gram length to list of n-grams.
+  ngrams = {i: [] for i in range(1, max_ngram_size+1)}
+  for gram_size in range(1, max_ngram_size+1):
+    for g in _window(grams, gram_size):
+      if _contiguous(g):
+        # Add an n-gram which spans these one-grams.
+        ngrams[gram_size].append(_Gram(g[0].begin, g[-1].end))
+
+  # Shuffle each list of n-grams.
+  for v in ngrams.values():
+    rng.shuffle(v)
+
+  # Create the weighting for n-gram length selection.
+  # Stored cummulatively for `random.choices` below.
+  cummulative_weights = list(
+      itertools.accumulate([1./n for n in range(1, max_ngram_size+1)]))
+
+  output_ngrams = []
+  # Keep a bitmask of which tokens have been masked.
+  masked_tokens = [False] * num_tokens
+  # Loop until we have enough masked tokens or there are no more candidate
+  # n-grams of any length.
+  # Each code path should ensure one or more elements from `ngrams` are removed
+  # to guarentee this loop terminates.
+  while (sum(masked_tokens) < max_masked_tokens and
+         sum(len(s) for s in ngrams.values())):
+    # Pick an n-gram size based on our weights.
+    sz = random.choices(range(1, max_ngram_size+1),
+                        cum_weights=cummulative_weights)[0]
+
+    # Ensure this size doesn't result in too many masked tokens.
+    # E.g., a two-gram contains _at least_ two tokens.
+    if sum(masked_tokens) + sz > max_masked_tokens:
+      # All n-grams of this length are too long and can be removed from
+      # consideration.
+      ngrams[sz].clear()
+      continue
 
-def create_masked_lm_predictions(tokens, masked_lm_prob,
-                                 max_predictions_per_seq, vocab_words, rng,
-                                 do_whole_word_mask):
-  """Creates the predictions for the masked LM objective."""
+    # All of the n-grams of this size have been used.
+    if not ngrams[sz]:
+      continue
+
+    # Choose a random n-gram of the given size.
+    gram = ngrams[sz].pop()
+    num_gram_tokens = gram.end-gram.begin
+
+    # Check if this would add too many tokens.
+    if num_gram_tokens + sum(masked_tokens) > max_masked_tokens:
+      continue
+
+    # Check if any of the tokens in this gram have already been masked.
+    if sum(masked_tokens[gram.begin:gram.end]):
+      continue
 
-  cand_indexes = []
-  for (i, token) in enumerate(tokens):
-    if token == "[CLS]" or token == "[SEP]":
+    # Found a usable n-gram!  Mark its tokens as masked and add it to return.
+    masked_tokens[gram.begin:gram.end] = [True] * (gram.end-gram.begin)
+    output_ngrams.append(gram)
+  return output_ngrams
+
+
+def _wordpieces_to_grams(tokens):
+  """Reconstitue grams (words) from `tokens`.
+
+  E.g.,
+     tokens: ['[CLS]', 'That', 'lit', '##tle', 'blue', 'tru', '##ck', '[SEP]']
+      grams: [          [1,2), [2,         4),  [4,5) , [5,       6)]
+
+  Arguments:
+    tokens: list of wordpieces
+  Returns:
+    List of _Grams representing spans of whole words
+    (without "[CLS]" and "[SEP]").
+  """
+  grams = []
+  gram_start_pos = None
+  for i, token in enumerate(tokens):
+    if gram_start_pos is not None and token.startswith("##"):
       continue
-    # Whole Word Masking means that if we mask all of the wordpieces
-    # corresponding to an original word. When a word has been split into
-    # WordPieces, the first token does not have any marker and any subsequence
-    # tokens are prefixed with ##. So whenever we see the ## token, we
-    # append it to the previous set of word indexes.
-    #
-    # Note that Whole Word Masking does *not* change the training code
-    # at all -- we still predict each WordPiece independently, softmaxed
-    # over the entire vocabulary.
-    if (do_whole_word_mask and len(cand_indexes) >= 1 and
-        token.startswith("##")):
-      cand_indexes[-1].append(i)
+    if gram_start_pos is not None:
+      grams.append(_Gram(gram_start_pos, i))
+    if token not in ["[CLS]", "[SEP]"]:
+      gram_start_pos = i
     else:
-      cand_indexes.append([i])
+      gram_start_pos = None
+  if gram_start_pos is not None:
+    grams.append(_Gram(gram_start_pos, len(tokens)))
+  return grams
 
-  rng.shuffle(cand_indexes)
 
-  output_tokens = list(tokens)
+def create_masked_lm_predictions(tokens, masked_lm_prob,
+                                 max_predictions_per_seq, vocab_words, rng,
+                                 do_whole_word_mask,
+                                 max_ngram_size=None):
+  """Creates the predictions for the masked LM objective."""
+  if do_whole_word_mask:
+    grams = _wordpieces_to_grams(tokens)
+  else:
+    # Here we consider each token to be a word to allow for sub-word masking.
+    if max_ngram_size:
+      raise ValueError("cannot use ngram masking without whole word masking")
+    grams = [_Gram(i, i+1) for i in range(0, len(tokens))
+             if tokens[i] not in ["[CLS]", "[SEP]"]]
 
   num_to_predict = min(max_predictions_per_seq,
                        max(1, int(round(len(tokens) * masked_lm_prob))))
-
+  # Generate masks.  If `max_ngram_size` in [0, None] it means we're doing
+  # whole word masking or token level masking.  Both of these can be treated
+  # as the `max_ngram_size=1` case.
+  masked_grams = _masking_ngrams(grams, max_ngram_size or 1,
+                                 num_to_predict, rng)
   masked_lms = []
-  covered_indexes = set()
-  for index_set in cand_indexes:
-    if len(masked_lms) >= num_to_predict:
-      break
-    # If adding a whole-word mask would exceed the maximum number of
-    # predictions, then just skip this candidate.
-    if len(masked_lms) + len(index_set) > num_to_predict:
-      continue
-    is_any_index_covered = False
-    for index in index_set:
-      if index in covered_indexes:
-        is_any_index_covered = True
-        break
-    if is_any_index_covered:
-      continue
-    for index in index_set:
-      covered_indexes.add(index)
-
-      masked_token = None
-      # 80% of the time, replace with [MASK]
+  output_tokens = list(tokens)
+  for gram in masked_grams:
+    # 80% of the time, replace all n-gram tokens with [MASK]
     if rng.random() < 0.8:
-        masked_token = "[MASK]"
+      replacement_action = lambda idx: "[MASK]"
     else:
-        # 10% of the time, keep original
+      # 10% of the time, keep all the original n-gram tokens.
       if rng.random() < 0.5:
-          masked_token = tokens[index]
-        # 10% of the time, replace with random word
+        replacement_action = lambda idx: tokens[idx]
+      # 10% of the time, replace each n-gram token with a random word.
       else:
-          masked_token = vocab_words[rng.randint(0, len(vocab_words) - 1)]
+        replacement_action = lambda idx: rng.choice(vocab_words)
 
-      output_tokens[index] = masked_token
+    for idx in range(gram.begin, gram.end):
+      output_tokens[idx] = replacement_action(idx)
+      masked_lms.append(MaskedLmInstance(index=idx, label=tokens[idx]))
 
-      masked_lms.append(MaskedLmInstance(index=index, label=tokens[index]))
   assert len(masked_lms) <= num_to_predict
   masked_lms = sorted(masked_lms, key=lambda x: x.index)
 
@@ -467,7 +642,7 @@ def main(_):
   instances = create_training_instances(
       input_files, tokenizer, FLAGS.max_seq_length, FLAGS.dupe_factor,
       FLAGS.short_seq_prob, FLAGS.masked_lm_prob, FLAGS.max_predictions_per_seq,
-      rng, FLAGS.do_whole_word_mask)
+      rng, FLAGS.do_whole_word_mask, FLAGS.max_ngram_size)
 
   output_files = FLAGS.output_file.split(",")
   logging.info("*** Writing to output files ***")

official/nlp/data/question_answering_dataloader.py

+# Lint as: python3
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# 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
+#
+#     http://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.
+# ==============================================================================
+"""Loads dataset for the question answering (e.g, SQuAD) task."""
+from typing import Mapping, Optional
+import dataclasses
+import tensorflow as tf
+
+from official.core import input_reader
+from official.modeling.hyperparams import config_definitions as cfg
+from official.nlp.data import data_loader_factory
+
+
+@dataclasses.dataclass
+class QADataConfig(cfg.DataConfig):
+  """Data config for question answering task (tasks/question_answering)."""
+  input_path: str = ''
+  global_batch_size: int = 48
+  is_training: bool = True
+  seq_length: int = 384
+  # Settings below are question answering specific.
+  version_2_with_negative: bool = False
+  # Settings below are only used for eval mode.
+  input_preprocessed_data_path: str = ''
+  doc_stride: int = 128
+  query_length: int = 64
+  vocab_file: str = ''
+  tokenization: str = 'WordPiece'  # WordPiece or SentencePiece
+  do_lower_case: bool = True
+
+
+@data_loader_factory.register_data_loader_cls(QADataConfig)
+class QuestionAnsweringDataLoader:
+  """A class to load dataset for sentence prediction (classification) task."""
+
+  def __init__(self, params):
+    self._params = params
+    self._seq_length = params.seq_length
+    self._is_training = params.is_training
+
+  def _decode(self, record: tf.Tensor):
+    """Decodes a serialized tf.Example."""
+    name_to_features = {
+        'input_ids': tf.io.FixedLenFeature([self._seq_length], tf.int64),
+        'input_mask': tf.io.FixedLenFeature([self._seq_length], tf.int64),
+        'segment_ids': tf.io.FixedLenFeature([self._seq_length], tf.int64),
+    }
+    if self._is_training:
+      name_to_features['start_positions'] = tf.io.FixedLenFeature([], tf.int64)
+      name_to_features['end_positions'] = tf.io.FixedLenFeature([], tf.int64)
+    else:
+      name_to_features['unique_ids'] = tf.io.FixedLenFeature([], tf.int64)
+    example = tf.io.parse_single_example(record, name_to_features)
+
+    # tf.Example only supports tf.int64, but the TPU only supports tf.int32.
+    # So cast all int64 to int32.
+    for name in example:
+      t = example[name]
+      if t.dtype == tf.int64:
+        t = tf.cast(t, tf.int32)
+      example[name] = t
+
+    return example
+
+  def _parse(self, record: Mapping[str, tf.Tensor]):
+    """Parses raw tensors into a dict of tensors to be consumed by the model."""
+    x, y = {}, {}
+    for name, tensor in record.items():
+      if name in ('start_positions', 'end_positions'):
+        y[name] = tensor
+      elif name == 'input_ids':
+        x['input_word_ids'] = tensor
+      elif name == 'segment_ids':
+        x['input_type_ids'] = tensor
+      else:
+        x[name] = tensor
+    return (x, y)
+
+  def load(self, input_context: Optional[tf.distribute.InputContext] = None):
+    """Returns a tf.dataset.Dataset."""
+    reader = input_reader.InputReader(
+        params=self._params, decoder_fn=self._decode, parser_fn=self._parse)
+    return reader.read(input_context)

official/nlp/data/sentence_prediction_dataloader.py

@@ -23,6 +23,9 @@ from official.modeling.hyperparams import config_definitions as cfg
 from official.nlp.data import data_loader_factory
 
 
+LABEL_TYPES_MAP = {'int': tf.int64, 'float': tf.float32}
+
+
 @dataclasses.dataclass
 class SentencePredictionDataConfig(cfg.DataConfig):
   """Data config for sentence prediction task (tasks/sentence_prediction)."""
@@ -30,6 +33,7 @@ class SentencePredictionDataConfig(cfg.DataConfig):
   global_batch_size: int = 32
   is_training: bool = True
   seq_length: int = 128
+  label_type: str = 'int'
 
 
 @data_loader_factory.register_data_loader_cls(SentencePredictionDataConfig)
@@ -42,11 +46,12 @@ class SentencePredictionDataLoader:
 
   def _decode(self, record: tf.Tensor):
     """Decodes a serialized tf.Example."""
+    label_type = LABEL_TYPES_MAP[self._params.label_type]
     name_to_features = {
         'input_ids': tf.io.FixedLenFeature([self._seq_length], tf.int64),
         'input_mask': tf.io.FixedLenFeature([self._seq_length], tf.int64),
         'segment_ids': tf.io.FixedLenFeature([self._seq_length], tf.int64),
-        'label_ids': tf.io.FixedLenFeature([], tf.int64),
+        'label_ids': tf.io.FixedLenFeature([], label_type),
     }
     example = tf.io.parse_single_example(record, name_to_features)
 

official/nlp/data/tagging_data_loader.py

@@ -28,6 +28,7 @@ class TaggingDataConfig(cfg.DataConfig):
   """Data config for tagging (tasks/tagging)."""
   is_training: bool = True
   seq_length: int = 128
+  include_sentence_id: bool = False
 
 
 @data_loader_factory.register_data_loader_cls(TaggingDataConfig)
@@ -37,6 +38,7 @@ class TaggingDataLoader:
   def __init__(self, params: TaggingDataConfig):
     self._params = params
     self._seq_length = params.seq_length
+    self._include_sentence_id = params.include_sentence_id
 
   def _decode(self, record: tf.Tensor):
     """Decodes a serialized tf.Example."""
@@ -46,6 +48,9 @@ class TaggingDataLoader:
         'segment_ids': tf.io.FixedLenFeature([self._seq_length], tf.int64),
         'label_ids': tf.io.FixedLenFeature([self._seq_length], tf.int64),
     }
+    if self._include_sentence_id:
+      name_to_features['sentence_id'] = tf.io.FixedLenFeature([], tf.int64)
+
     example = tf.io.parse_single_example(record, name_to_features)
 
     # tf.Example only supports tf.int64, but the TPU only supports tf.int32.
@@ -65,6 +70,8 @@ class TaggingDataLoader:
         'input_mask': record['input_mask'],
         'input_type_ids': record['segment_ids']
     }
+    if self._include_sentence_id:
+      x['sentence_id'] = record['sentence_id']
     y = record['label_ids']
     return (x, y)
 

official/nlp/modeling/README.md

 # NLP Modeling Library
 
-This libary provides a set of Keras primitives (Layers, Networks, and Models)
+This library provides a set of Keras primitives (Layers, Networks, and Models)
 that can be assembled into transformer-based models. They are
 flexible, validated, interoperable, and both TF1 and TF2 compatible.
 
@@ -16,6 +16,11 @@ standardized configuration.
 
 * [`losses`](losses) contains common loss computation used in NLP tasks.
 
+Please see the colab
+[nlp_modeling_library_intro.ipynb]
+(https://colab.sandbox.google.com/github/tensorflow/models/blob/master/official/colab/nlp/nlp_modeling_library_intro.ipynb)
+for how to build transformer-based NLP models using above primitives.
+
 Besides the pre-defined primitives, it also provides scaffold classes to allow
 easy experimentation with noval achitectures, e.g., you don’t need to fork a whole Transformer object to try a different kind of attention primitive, for instance.
 
@@ -33,11 +38,9 @@ embedding subnetwork (which will replace the standard embedding logic) and/or a
 custom hidden layer (which will replace the Transformer instantiation in the
 encoder).
 
-BERT and ALBERT models in this repo are implemented using this library. Code examples can be found in the corresponding model folder.
-
-
-
-
-
-
+Please see the colab
+[customize_encoder.ipynb]
+(https://colab.sandbox.google.com/github/tensorflow/models/blob/master/official/colab/nlp/customize_encoder.ipynb)
+for how to use scaffold classes to build noval achitectures.
 
+BERT and ALBERT models in this repo are implemented using this library. Code examples can be found in the corresponding model folder.

official/nlp/modeling/layers/README.md

@@ -3,11 +3,6 @@
 Layers are the fundamental building blocks for NLP models. They can be used to
 assemble new layers, networks, or models.
 
-*   [DenseEinsum](dense_einsum.py) implements a feedforward network using
-    tf.einsum. This layer contains the einsum op, the associated weight, and the
-    logic required to generate the einsum expression for the given
-    initialization parameters.
-
 *   [MultiHeadAttention](attention.py) implements an optionally masked attention
     between query, key, value tensors as described in
     ["Attention Is All You Need"](https://arxiv.org/abs/1706.03762). If

official/nlp/modeling/layers/attention.py

@@ -33,7 +33,7 @@ EinsumDense = tf.keras.layers.experimental.EinsumDense
 _CHR_IDX = string.ascii_lowercase
 
 
-def _build_attention_equation(qkv_rank, attn_axes):
+def _build_attention_equation(rank, attn_axes):
   """Builds einsum equations for the attention computation.
 
   Query, key, value inputs after projection are expected to have the shape as:
@@ -50,19 +50,19 @@ def _build_attention_equation(qkv_rank, attn_axes):
   <query attention dims>, num_heads, channels)
 
   Args:
-    qkv_rank: the rank of query, key, value tensors.
+    rank: the rank of query, key, value tensors.
     attn_axes: a list/tuple of axes, [1, rank), that will do attention.
 
   Returns:
     Einsum equations.
   """
-  target_notation = _CHR_IDX[:qkv_rank]
+  target_notation = _CHR_IDX[:rank]
   # `batch_dims` includes the head dim.
-  batch_dims = tuple(np.delete(range(qkv_rank), attn_axes + (qkv_rank - 1,)))
-  letter_offset = qkv_rank
+  batch_dims = tuple(np.delete(range(rank), attn_axes + (rank - 1,)))
+  letter_offset = rank
   source_notation = ""
-  for i in range(qkv_rank):
-    if i in batch_dims or i == qkv_rank - 1:
+  for i in range(rank):
+    if i in batch_dims or i == rank - 1:
       source_notation += target_notation[i]
     else:
       source_notation += _CHR_IDX[letter_offset]
@@ -167,8 +167,8 @@ class MultiHeadAttention(tf.keras.layers.Layer):
       sequence dims. If not specified, projects back to the key feature dim.
     attention_axes: axes over which the attention is applied. `None` means
       attention over all axes, but batch, heads, and features.
-    return_attention_scores: bool, if `True`, returns the multi-head
-      attention scores as an additional output argument.
+    return_attention_scores: bool, if `True`, returns the multi-head attention
+      scores as an additional output argument.
     kernel_initializer: Initializer for dense layer kernels.
     bias_initializer: Initializer for dense layer biases.
     kernel_regularizer: Regularizer for dense layer kernels.
@@ -176,6 +176,13 @@ class MultiHeadAttention(tf.keras.layers.Layer):
     activity_regularizer: Regularizer for dense layer activity.
     kernel_constraint: Constraint for dense layer kernels.
     bias_constraint: Constraint for dense layer kernels.
+  Call args:
+    query: Query `Tensor` of shape `[B, T, dim]`.
+    value: Value `Tensor` of shape `[B, S, dim]`.
+    key: Optional key `Tensor` of shape `[B, S, dim]`. If not given, will use
+      `value` for both `key` and `value`, which is the most common case.
+    attention_mask: a boolean mask of shape `[B, T, S]`, that prevents attention
+      to certain positions.
   """
 
   def __init__(self,
@@ -214,6 +221,7 @@ class MultiHeadAttention(tf.keras.layers.Layer):
       self._attention_axes = (attention_axes,)
     else:
       self._attention_axes = attention_axes
+    self._built_from_signature = False
 
   def get_config(self):
     config = {
@@ -251,17 +259,31 @@ class MultiHeadAttention(tf.keras.layers.Layer):
     base_config = super(MultiHeadAttention, self).get_config()
     return dict(list(base_config.items()) + list(config.items()))
 
-  def build(self, input_shape):
-    inputs_len = len(input_shape)
-    if inputs_len > 3 or inputs_len < 2:
-      raise ValueError(
-          "Expects inputs list of length 2 or 3, namely [query, value] or "
-          "[query, value, key]. "
-          "Given length: %d" % inputs_len)
-    tensor_shapes = tf.nest.map_structure(tf.TensorShape, input_shape)
-    query_shape = tensor_shapes[0]
-    value_shape = tensor_shapes[1]
-    key_shape = tensor_shapes[2] if inputs_len == 3 else value_shape
+  def _build_from_signature(self, query, value, key=None):
+    """Builds layers and variables.
+
+    Once the method is called, self._built_from_signature will be set to True.
+
+    Args:
+      query: query tensor or TensorShape.
+      value: value tensor or TensorShape.
+      key: key tensor or TensorShape.
+    """
+    self._built_from_signature = True
+    if hasattr(query, "shape"):
+      query_shape = tf.TensorShape(query.shape)
+    else:
+      query_shape = query
+    if hasattr(value, "shape"):
+      value_shape = tf.TensorShape(value.shape)
+    else:
+      value_shape = value
+    if key is None:
+      key_shape = value_shape
+    elif hasattr(key, "shape"):
+      key_shape = tf.TensorShape(key.shape)
+    else:
+      key_shape = key
 
     common_kwargs = dict(
         kernel_initializer=self._kernel_initializer,
@@ -271,7 +293,7 @@ class MultiHeadAttention(tf.keras.layers.Layer):
         activity_regularizer=self._activity_regularizer,
         kernel_constraint=self._kernel_constraint,
         bias_constraint=self._bias_constraint)
-
+    with tf.init_scope():
       free_dims = query_shape.rank - 1
       einsum_equation, bias_axes, output_rank = _build_proj_equation(
           free_dims, bound_dims=1, output_dims=2)
@@ -302,9 +324,9 @@ class MultiHeadAttention(tf.keras.layers.Layer):
           **common_kwargs)
 
       # Builds the attention computations for multi-head dot product attention.
-    # These computations could be wrapped into the keras attention layer once it
-    # support mult-head einsum computations.
-    self._build_attention(output_rank)
+      # These computations could be wrapped into the keras attention layer once
+      # it support mult-head einsum computations.
+      self.build_attention(output_rank)
       if self._output_shape:
         if not isinstance(self._output_shape, collections.abc.Sized):
           output_shape = [self._output_shape]
@@ -320,35 +342,30 @@ class MultiHeadAttention(tf.keras.layers.Layer):
           bias_axes=bias_axes if self._use_bias else None,
           name="attention_output",
           **common_kwargs)
-    super(MultiHeadAttention, self).build(input_shape)
 
-  def _build_attention(self, qkv_rank):
+  def build_attention(self, rank):
     """Builds multi-head dot-product attention computations.
 
-    This function builds attributes necessary for `_compute_attention` to
+    This function builds attributes necessary for `compute_attention` to
     costomize attention computation to replace the default dot-product
     attention.
 
     Args:
-      qkv_rank: the rank of query, key, value tensors.
+      rank: the rank of query, key, value tensors.
     """
     if self._attention_axes is None:
-      self._attention_axes = tuple(range(1, qkv_rank - 2))
+      self._attention_axes = tuple(range(1, rank - 2))
     else:
       self._attention_axes = tuple(self._attention_axes)
     self._dot_product_equation, self._combine_equation, attn_scores_rank = (
-        _build_attention_equation(qkv_rank, attn_axes=self._attention_axes))
+        _build_attention_equation(rank, attn_axes=self._attention_axes))
     norm_axes = tuple(
         range(attn_scores_rank - len(self._attention_axes), attn_scores_rank))
     self._masked_softmax = masked_softmax.MaskedSoftmax(
         mask_expansion_axes=[1], normalization_axes=norm_axes)
     self._dropout_layer = tf.keras.layers.Dropout(rate=self._dropout)
 
-  def _compute_attention(self,
-                         query_tensor,
-                         key_tensor,
-                         value_tensor,
-                         attention_mask=None):
+  def compute_attention(self, query, key, value, attention_mask=None):
     """Applies Dot-product attention with query, key, value tensors.
 
     This function defines the computation inside `call` with projected
@@ -356,9 +373,9 @@ class MultiHeadAttention(tf.keras.layers.Layer):
     attention implementation.
 
     Args:
-      query_tensor: Projected query `Tensor` of shape `[B, T, N, key_size]`.
-      key_tensor: Projected key `Tensor` of shape `[B, T, N, key_size]`.
-      value_tensor: Projected value `Tensor` of shape `[B, T, N, value_size]`.
+      query: Projected query `Tensor` of shape `[B, T, N, key_size]`.
+      key: Projected key `Tensor` of shape `[B, T, N, key_size]`.
+      value: Projected value `Tensor` of shape `[B, T, N, value_size]`.
       attention_mask: a boolean mask of shape `[B, T, S]`, that prevents
         attention to certain positions.
 
@@ -366,12 +383,14 @@ class MultiHeadAttention(tf.keras.layers.Layer):
       attention_output: Multi-headed outputs of attention computation.
       attention_scores: Multi-headed attention weights.
     """
+    # Note: Applying scalar multiply at the smaller end of einsum improves
+    # XLA performance, but may introduce slight numeric differences in
+    # the Transformer attention head.
+    query = tf.multiply(query, 1.0 / math.sqrt(float(self._key_size)))
+
     # Take the dot product between "query" and "key" to get the raw
     # attention scores.
-    attention_scores = tf.einsum(self._dot_product_equation, key_tensor,
-                                 query_tensor)
-    attention_scores = tf.multiply(attention_scores,
-                                   1.0 / math.sqrt(float(self._key_size)))
+    attention_scores = tf.einsum(self._dot_product_equation, key, query)
 
     # Normalize the attention scores to probabilities.
     # `attention_scores` = [B, N, T, S]
@@ -383,10 +402,10 @@ class MultiHeadAttention(tf.keras.layers.Layer):
 
     # `context_layer` = [B, T, N, H]
     attention_output = tf.einsum(self._combine_equation,
-                                 attention_scores_dropout, value_tensor)
+                                 attention_scores_dropout, value)
     return attention_output, attention_scores
 
-  def call(self, inputs, attention_mask=None):
+  def call(self, query, value, key=None, attention_mask=None):
     """Implements the forward pass.
 
     Size glossary:
@@ -399,11 +418,10 @@ class MultiHeadAttention(tf.keras.layers.Layer):
       * Value (source) attention axes shape (S), the rank must match the target.
 
     Args:
-      inputs: List of the following tensors:
-        * query: Query `Tensor` of shape `[B, T, dim]`.
-        * value: Value `Tensor` of shape `[B, S, dim]`.
-        * key: Optional key `Tensor` of shape `[B, S, dim]`. If not given, will
-          use `value` for both `key` and `value`, which is the most common case.
+      query: Query `Tensor` of shape `[B, T, dim]`.
+      value: Value `Tensor` of shape `[B, S, dim]`.
+      key: Optional key `Tensor` of shape `[B, S, dim]`. If not given, will use
+        `value` for both `key` and `value`, which is the most common case.
       attention_mask: a boolean mask of shape `[B, T, S]`, that prevents
         attention to certain positions.
 
@@ -416,29 +434,24 @@ class MultiHeadAttention(tf.keras.layers.Layer):
       attention
         axes.
     """
-    inputs_len = len(inputs)
-    if inputs_len > 3 or inputs_len < 2:
-      raise ValueError(
-          "Expects inputs list of length 2 or 3, namely [query, value] or "
-          "[query, value, key]. "
-          "Given length: %d" % inputs_len)
-    query = inputs[0]
-    value = inputs[1]
-    key = inputs[2] if inputs_len == 3 else value
+    if not self._built_from_signature:
+      self._build_from_signature(query=query, value=value, key=key)
+    if key is None:
+      key = value
 
     #   N = `num_attention_heads`
     #   H = `size_per_head`
-    # `query_tensor` = [B, T, N ,H]
-    query_tensor = self._query_dense(query)
+    # `query` = [B, T, N ,H]
+    query = self._query_dense(query)
 
-    # `key_tensor` = [B, S, N, H]
-    key_tensor = self._key_dense(key)
+    # `key` = [B, S, N, H]
+    key = self._key_dense(key)
 
-    # `value_tensor` = [B, S, N, H]
-    value_tensor = self._value_dense(value)
+    # `value` = [B, S, N, H]
+    value = self._value_dense(value)
 
-    attention_output, attention_scores = self._compute_attention(
-        query_tensor, key_tensor, value_tensor, attention_mask)
+    attention_output, attention_scores = self.compute_attention(
+        query, key, value, attention_mask)
     attention_output = self._output_dense(attention_output)
 
     if self._return_attention_scores:
@@ -453,40 +466,42 @@ class CachedAttention(MultiHeadAttention):
   Arguments are the same as `MultiHeadAttention` layer.
   """
 
-  def _update_cache(self, key_tensor, value_tensor, cache, decode_loop_step):
+  def _update_cache(self, key, value, cache, decode_loop_step):
     """Updates cache states and gets full-length key/value tensors."""
     # Combines cached keys and values with new keys and values.
     if decode_loop_step is not None:
       # TPU special case.
       key_seq_dim = cache["key"].shape.as_list()[1]
       indices = tf.reshape(
-          tf.one_hot(decode_loop_step, key_seq_dim, dtype=key_tensor.dtype),
+          tf.one_hot(decode_loop_step, key_seq_dim, dtype=key.dtype),
           [1, key_seq_dim, 1, 1])
-      key_tensor = cache["key"] + key_tensor * indices
+      key = cache["key"] + key * indices
       value_seq_dim = cache["value"].shape.as_list()[1]
       indices = tf.reshape(
-          tf.one_hot(decode_loop_step, value_seq_dim, dtype=value_tensor.dtype),
+          tf.one_hot(decode_loop_step, value_seq_dim, dtype=value.dtype),
           [1, value_seq_dim, 1, 1])
-      value_tensor = cache["value"] + value_tensor * indices
+      value = cache["value"] + value * indices
     else:
-      key_tensor = tf.concat(
-          [tf.cast(cache["key"], key_tensor.dtype), key_tensor], axis=1)
-      value_tensor = tf.concat(
-          [tf.cast(cache["value"], value_tensor.dtype), value_tensor], axis=1)
+      key = tf.concat([tf.cast(cache["key"], key.dtype), key], axis=1)
+      value = tf.concat([tf.cast(cache["value"], value.dtype), value], axis=1)
 
     # Update cache
-    cache["key"] = key_tensor
-    cache["value"] = value_tensor
+    cache["key"] = key
+    cache["value"] = value
 
-    return key_tensor, value_tensor
+    return key, value
 
   def call(self,
-           inputs,
+           query,
+           value,
+           key=None,
            attention_mask=None,
            cache=None,
            decode_loop_step=None):
-    from_tensor = inputs[0]
-    to_tensor = inputs[1]
+    if not self._built_from_signature:
+      self._build_from_signature(query=query, value=value, key=key)
+    if key is None:
+      key = value
 
     # Scalar dimensions referenced here:
     #   B = batch size (number of sequences)
@@ -494,23 +509,21 @@ class CachedAttention(MultiHeadAttention):
     #   T = `to_tensor` sequence length
     #   N = `num_attention_heads`
     #   H = `size_per_head`
-    # `query_tensor` = [B, F, N ,H]
-    query_tensor = self._query_dense(from_tensor)
+    # `query` = [B, F, N ,H]
+    query = self._query_dense(query)
 
-    # `key_tensor` = [B, T, N, H]
-    key_tensor = self._key_dense(to_tensor)
+    # `key` = [B, T, N, H]
+    key = self._key_dense(key)
 
-    # `value_tensor` = [B, T, N, H]
-    value_tensor = self._value_dense(to_tensor)
+    # `value` = [B, T, N, H]
+    value = self._value_dense(value)
 
     if cache:
-      key_tensor, value_tensor = self._update_cache(key_tensor, value_tensor,
-                                                    cache, decode_loop_step)
+      key, value = self._update_cache(key, value, cache, decode_loop_step)
 
     # Take the dot product between "query" and "key" to get the raw
     # attention scores.
-    attention_scores = tf.einsum(self._dot_product_equation, key_tensor,
-                                 query_tensor)
+    attention_scores = tf.einsum(self._dot_product_equation, key, query)
     attention_scores = tf.multiply(attention_scores,
                                    1.0 / math.sqrt(float(self._key_size)))
 
@@ -523,7 +536,7 @@ class CachedAttention(MultiHeadAttention):
     attention_scores = self._dropout_layer(attention_scores)
     # `context_layer` = [B, F, N, H]
     attention_output = tf.einsum(self._combine_equation, attention_scores,
-                                 value_tensor)
+                                 value)
     attention_output = self._output_dense(attention_output)
     if self._return_attention_scores:
       return attention_output, attention_scores, cache

official/nlp/modeling/layers/attention_test.py

@@ -45,7 +45,7 @@ class MultiHeadAttentionTest(keras_parameterized.TestCase):
     # Create a 3-dimensional input (the first dimension is implicit).
     query = tf.keras.Input(shape=(40, 80))
     value = tf.keras.Input(shape=(20, 80))
-    output = test_layer([query, value])
+    output = test_layer(query=query, value=value)
     self.assertEqual(output.shape.as_list(), [None] + output_dims)
 
   def test_non_masked_self_attention(self):
@@ -53,7 +53,7 @@ class MultiHeadAttentionTest(keras_parameterized.TestCase):
     test_layer = attention.MultiHeadAttention(num_heads=12, key_size=64)
     # Create a 3-dimensional input (the first dimension is implicit).
     query = tf.keras.Input(shape=(40, 80))
-    output = test_layer([query, query])
+    output = test_layer(query, query)
     self.assertEqual(output.shape.as_list(), [None, 40, 80])
 
   def test_attention_scores(self):
@@ -62,7 +62,7 @@ class MultiHeadAttentionTest(keras_parameterized.TestCase):
         num_heads=12, key_size=64, return_attention_scores=True)
     # Create a 3-dimensional input (the first dimension is implicit).
     query = tf.keras.Input(shape=(40, 80))
-    output, coef = test_layer([query, query])
+    output, coef = test_layer(query, query)
     self.assertEqual(output.shape.as_list(), [None, 40, 80])
     self.assertEqual(coef.shape.as_list(), [None, 12, 40, 40])
 
@@ -76,7 +76,7 @@ class MultiHeadAttentionTest(keras_parameterized.TestCase):
     query = tf.keras.Input(shape=(4, 8))
     value = tf.keras.Input(shape=(2, 8))
     mask_tensor = tf.keras.Input(shape=(4, 2))
-    output = test_layer([query, value], mask_tensor)
+    output = test_layer(query=query, value=value, attention_mask=mask_tensor)
 
     # Create a model containing the test layer.
     model = tf.keras.Model([query, value, mask_tensor], output)
@@ -100,7 +100,7 @@ class MultiHeadAttentionTest(keras_parameterized.TestCase):
 
     # Tests the layer with three inputs: Q, K, V.
     key = tf.keras.Input(shape=(2, 8))
-    output = test_layer([query, value, key], mask_tensor)
+    output = test_layer(query, value=value, key=key, attention_mask=mask_tensor)
     model = tf.keras.Model([query, value, key, mask_tensor], output)
 
     masked_output_data = model.predict([from_data, to_data, to_data, mask_data])
@@ -125,7 +125,7 @@ class MultiHeadAttentionTest(keras_parameterized.TestCase):
         kernel_initializer=tf.keras.initializers.TruncatedNormal(stddev=0.02))
     # Create a 3-dimensional input (the first dimension is implicit).
     query = tf.keras.Input(shape=(40, 80))
-    output = test_layer([query, query])
+    output = test_layer(query, query)
     self.assertEqual(output.shape.as_list(), [None, 40, 80])
 
   @parameterized.named_parameters(
@@ -147,11 +147,12 @@ class MultiHeadAttentionTest(keras_parameterized.TestCase):
     # Invoke the data with a random set of mask data. This should mask at least
     # one element.
     mask_data = np.random.randint(2, size=mask_shape).astype("bool")
-    output = test_layer([query, value], mask_data)
+    output = test_layer(query=query, value=value, attention_mask=mask_data)
 
     # Invoke the same data, but with a null mask (where no elements are masked).
     null_mask_data = np.ones(mask_shape)
-    unmasked_output = test_layer([query, value], null_mask_data)
+    unmasked_output = test_layer(
+        query=query, value=value, attention_mask=null_mask_data)
     # Because one data is masked and one is not, the outputs should not be the
     # same.
     self.assertNotAllClose(output, unmasked_output)
@@ -180,7 +181,7 @@ class AttentionSubclassTest(keras_parameterized.TestCase):
         key_size=64)
     # Create a 3-dimensional input (the first dimension is implicit).
     query = tf.keras.Input(shape=(40, 80))
-    output = test_layer([query, query])
+    output = test_layer(query, query)
     self.assertEqual(output.shape.as_list(), [None, 40, 80])
 
 
@@ -216,12 +217,14 @@ class CachedAttentionTest(keras_parameterized.TestCase):
     # one element.
     mask_data = np.random.randint(
         2, size=(batch_size, from_seq_length, from_seq_length))
-    masked_output_data, cache = layer([from_data, from_data], mask_data, cache)
+    masked_output_data, cache = layer(
+        query=from_data, value=from_data, attention_mask=mask_data, cache=cache)
     self.assertEqual(masked_output_data.shape, (3, 4, 8))
     self.assertEqual(cache["value"].shape, (3, 4, 2, 2))
 
     # Tests inputs without cache.
-    masked_output_data, cache = layer([from_data, from_data, mask_data])
+    masked_output_data, cache = layer(
+        query=from_data, value=from_data, attention_mask=mask_data)
     self.assertEqual(masked_output_data.shape, (3, 4, 8))
     self.assertIsNone(cache)
 
@@ -243,9 +246,11 @@ class CachedAttentionTest(keras_parameterized.TestCase):
     mask_data = np.random.randint(
         2, size=(batch_size, from_seq_length, from_seq_length), dtype=np.int32)
     # Testing the invocation directly as Keras cannot consume inputs correctly.
-    masked_output_data, cache = layer([from_data, from_data],
-                                      mask_data,
-                                      cache,
+    masked_output_data, cache = layer(
+        query=from_data,
+        value=from_data,
+        attention_mask=mask_data,
+        cache=cache,
         decode_loop_step=decode_loop_step)
     self.assertEqual(masked_output_data.shape, (3, 4, 8))
     self.assertEqual(cache["value"].shape, (3, 4, 2, 2))