def _tokenize_example(document, summary):
tokenizer = tft.SentencepieceTokenizer(
model=tf.io.gfile.GFile(vocab_model_file, "rb").read())
if substitute_newline:
document = tf.strings.regex_replace(document, "\n",
substitute_newline)
# Remove space before special tokens.
document = tf.strings.regex_replace(document, r" ([<\[]\S+[>\]])",
b"\\1")
document_ids = tokenizer.tokenize(document)
if isinstance(document_ids, tf.RaggedTensor):
document_ids = document_ids.to_tensor(0)
document_ids = document_ids[:max_encoder_length]
# Remove newline optionally
if substitute_newline:
summary = tf.strings.regex_replace(summary, "\n",
substitute_newline)
# Remove space before special tokens.
summary = tf.strings.regex_replace(summary, r" ([<\[]\S+[>\]])",
b"\\1")
summary_ids = tokenizer.tokenize(summary)
# Add [EOS] (1) special tokens.
suffix = tf.constant([1])
summary_ids = tf.concat([summary_ids, suffix], axis=0)
if isinstance(summary_ids, tf.RaggedTensor):
summary_ids = summary_ids.to_tensor(0)
summary_ids = summary_ids[:max_decoder_length]
return document_ids, summary_ids
def input_fn():
# text input
text = tf.compat.v1.placeholder(tf.string, [batch_size],
name="input_text")
# text tokenize
tokenizer = tft.SentencepieceTokenizer(
model=tf.io.gfile.GFile(vocab_model_file, "rb").read())
if substitute_newline:
text = tf.strings.regex_replace(text, "\n", substitute_newline)
# Remove space before special tokens.
text = tf.strings.regex_replace(text, r" ([<\[]\S+[>\]])", b"\\1")
ids = tokenizer.tokenize(text)
if isinstance(ids, tf.RaggedTensor):
ids = ids.to_tensor(0)
# text padding: Pad only if necessary and reshape properly
padded_ids = dynamic_padding(ids, max_encoder_length)
ids = tf.slice(padded_ids, [0, 0], [batch_size, max_encoder_length])
receiver_tensors = {"input": text}
features = {"input_ids": tf.cast(ids, tf.int32, name="input_ids")}
return tf.estimator.export.ServingInputReceiver(
features=features, receiver_tensors=receiver_tensors)
def input_fn():
# text input
text = tf.compat.v1.placeholder(tf.string, [batch_size], name="input_text")
# text tokenize
tokenizer = tft.SentencepieceTokenizer(
model=tf.io.gfile.GFile(vocab_model_file, "rb").read())
if substitute_newline:
text = tf.strings.regex_replace(text, "\n", substitute_newline)
ids = tokenizer.tokenize(text)
ids = ids[:, :max_encoder_length - 2]
# Add [CLS] and [SEP] special tokens.
prefix = tf.repeat(tf.constant([[65]]), batch_size, axis=0)
suffix = tf.repeat(tf.constant([[66]]), batch_size, axis=0)
ids = tf.concat([prefix, ids, suffix], axis=1)
if isinstance(ids, tf.RaggedTensor):
ids = ids.to_tensor(0)
# text padding: Pad only if necessary and reshape properly
padded_ids = dynamic_padding(ids, max_encoder_length)
ids = tf.slice(padded_ids, [0, 0], [batch_size, max_encoder_length])
receiver_tensors = {"input": text}
features = {"input_ids": tf.cast(ids, tf.int32, name="input_ids")}
return tf.estimator.export.ServingInputReceiver(
features=features, receiver_tensors=receiver_tensors)
def _tokenize_example(context, question):
tokenizer = tft.SentencepieceTokenizer(
model=tf.io.gfile.GFile(vocab_model_file, "rb").read())
if substitute_newline:
context = tf.strings.regex_replace(context, "\n", substitute_newline)
# Remove space before special tokens.
context = tf.strings.regex_replace(context, r" ([<\[]\S+[>\]])", b"\\1")
context_ids = tokenizer.tokenize(context)
if isinstance(context_ids, tf.RaggedTensor):
context_ids = context_ids.to_tensor(0)
context_ids = context_ids[:max_encoder_length]
# Remove newline optionally
if substitute_newline:
question = tf.strings.regex_replace(question, "\n", substitute_newline)
# Remove space before special tokens.
question = tf.strings.regex_replace(question, r" ([<\[]\S+[>\]])", b"\\1")
question_ids = tokenizer.tokenize(question)
# Add [EOS] (1) special tokens.
suffix = tf.constant([1])
question_ids = tf.concat([question_ids, suffix], axis=0)
if isinstance(question_ids, tf.RaggedTensor):
question_ids = question_ids.to_tensor(0)
question_ids = question_ids[:max_decoder_length]
return context_ids, question_ids
def c4_preprocess(dataset, training, max_target_length=-1,
tokenization=None, spm_path=None):
"""Pre-processing function for C4 dataset."""
del training
def unicode_decode_chars(features, targets):
targets = tf.strings.unicode_decode(features['text'], 'UTF-8')
targets = tf.cast(targets, tf.int64)
features['targets'] = targets
features['inputs'] = targets
return (features, targets)
def spc_tokenize(tokenizer, features, targets):
del targets
tokenized_text = tokenizer.tokenize(features['text'])
features['targets'] = tf.cast(tokenized_text, tf.int64)
features['inputs'] = features['targets']
return features, features['targets']
if tokenization == 'spc':
spm_path = spm_path or t5_utils.DEFAULT_SPM_PATH
with tf.compat.v1.gfile.GFile(spm_path, 'rb') as f:
spc_model = f.read()
tokenizer = tf_text.SentencepieceTokenizer(model=spc_model)
dataset = dataset.map(functools.partial(spc_tokenize, tokenizer))
else:
dataset = dataset.map(unicode_decode_chars)
def target_right_length(_, target):
return tf.less(tf.shape(target)[0], max_target_length + 1)
if max_target_length > 0:
dataset = dataset.filter(target_right_length)
return dataset
def do_masking(example):
if "tfds://" == data_dir[:7]:
text = example["text"]
else:
text = example
print(text)
tokenizer = tft.SentencepieceTokenizer(
model=tf.io.gfile.GFile(vocab_model_file, "rb").read())
if substitute_newline:
text = tf.strings.regex_replace(text, "\n", substitute_newline)
subtokens = tokenizer.tokenize(text)
(subtokens, masked_lm_positions, masked_lm_ids,
masked_lm_weights) = tf.compat.v1.py_func(
numpy_masking, [subtokens],
[tf.int32, tf.int32, tf.int32, tf.float32],
stateful=False)
features = {
"input_ids": subtokens,
"segment_ids": tf.zeros_like(subtokens),
"masked_lm_positions": masked_lm_positions,
"masked_lm_ids": masked_lm_ids,
"masked_lm_weights": masked_lm_weights,
"next_sentence_labels": tf.zeros([1], dtype=tf.int64),
}
return features
def setup(self):
self.tokenizer = tft.SentencepieceTokenizer(
model=tf.io.gfile.GFile(self.vocab_model_file, "rb").read())
self.sentence_tokenizer = nltk.load(SENTENCE_TOKENIZER_PATH)
self.delimiter_range_pair = rendering_utils.get_default_delimiter_range_pair(
task=self.task,
delimiter_type=self.delimiter_type,
)
def load_sentencepiece_tokenizer(
model_path, add_bos=False, add_eos=True, reverse=False):
"""Load a tf-text SentencePiece tokenizer from given model filepath."""
with tf.io.gfile.GFile(model_path, 'rb') as model_fp:
sp_model = model_fp.read()
sp_tokenizer = tftxt.SentencepieceTokenizer(
model=sp_model, add_bos=add_bos, add_eos=add_eos, reverse=reverse)
return sp_tokenizer
def __init__(self, params: WMTDataConfig):
self._params = params
self._max_seq_length = params.max_seq_length
self._static_batch = params.static_batch
self._global_batch_size = params.global_batch_size
if self._params.transform_and_batch:
self._tokenizer = tftxt.SentencepieceTokenizer(
model=tf.io.gfile.GFile(params.sentencepiece_model_path,
'rb').read(),
add_eos=True)
def __post_init__(self):
tokenizer = tensorflow_text.SentencepieceTokenizer(
model=tf.io.gfile.GFile(self.vocab_path, 'rb').read(),
add_eos=True)
eos_token = tokenizer.string_to_id('</s>')
# Work-around for frozen dataclasses:
# https://stackoverflow.com/questions/53756788
object.__setattr__(self, 'eos_token', eos_token)
object.__setattr__(self, '_tokenizer', tokenizer)
def __init__(self, model=None, model_path=None):
super(SentencepieceTokenizer, self).__init__()
if model_path:
self.proto = open(model_path, "rb").read()
if model:
if isinstance(model, str):
asc_str = model.encode("ascii")
self.proto = base64.decodebytes(asc_str)
if isinstance(model, bytes):
self.proto = model
self.tokenizer = text.SentencepieceTokenizer(self.proto)
def __init__(self, params):
super().__init__(params)
p = self.params
with tf.io.gfile.GFile(p.spm_model, 'rb') as f:
self._tokenizer = tf_text.SentencepieceTokenizer(
model=f.read(),
out_type=tf.int32,
nbest_size=p.nbest_size,
alpha=p.alpha,
reverse=False,
add_bos=False,
add_eos=p.append_eos)
def __init__(self, params, model: tf.keras.Model, inference_step=None):
super().__init__(params, model, inference_step)
self._sp_tokenizer = tf_text.SentencepieceTokenizer(
model=tf.io.gfile.GFile(params.sentencepiece_model_path,
"rb").read(),
add_eos=True)
try:
empty_str_tokenized = self._sp_tokenizer.tokenize("").numpy()
except tf.errors.InternalError:
raise ValueError(
"EOS token not in tokenizer vocab."
"Please make sure the tokenizer generates a single token for an "
"empty string.")
self._eos_id = empty_str_tokenized.item()
def test_tftext_sentencepiece_tokenizer_bos_eos(self):
"""Check that the new tokenizer produces the same result that the tftext one with bos and eos."""
tftext_sp = tensorflow_text.SentencepieceTokenizer(
self.sentencepiece_model, add_bos=True, add_eos=True)
opt_sp = sentencepiece_tokenizer.SentencepieceTokenizer(
self.sentencepiece_model, add_bos=True, add_eos=True)
input_text = [
u" ", u"to be or not to be", u"ignored by length text1",
u"ignored by length text2"
]
tftext_tokenized = tftext_sp.tokenize(input_text)
opt_tokenized = opt_sp.tokenize(input_text)
self.assertAllEqual(tftext_tokenized, opt_tokenized)
def decode(ids: tf.Tensor, vocab_filename: str, encoder_type: str):
"""DecodeOp."""
if encoder_type not in ["sentencepiece", "sentencepiece_newline"]:
raise ValueError("Unsupported encoder type: %s" % encoder_type)
sp_model = tf.gfile.GFile(vocab_filename, "rb").read()
tokenizer = tf_text.SentencepieceTokenizer(model=sp_model)
ids = tf.where(ids > 1 + _SHIFT_RESERVED_TOKENS, ids - _SHIFT_RESERVED_TOKENS,
ids)
ids = tf.cast(ids, tf.int32)
text = tokenizer.detokenize(ids)
text = tf.reshape(text, [-1])
if encoder_type == "sentencepiece_newline":
text = tf.strings.regex_replace(text, _NEWLINE_SYMBOL, "\n")
return text
def test_tftext_sentencepiece_detokenizer(self):
"""Check that the new tokenizer produces the same result that the tftext one."""
tftext_sp = tensorflow_text.SentencepieceTokenizer(
self.sentencepiece_model)
opt_sp = sentencepiece_tokenizer.SentencepieceTokenizer(
self.sentencepiece_model)
input_text = [
u" ", u"to be or not to be", u"ignored by length text1",
u"ignored by length text2"
]
tftext_tokenized = tftext_sp.tokenize(input_text)
# Check detokenizer
tftext_detokenized = tftext_sp.detokenize(tftext_tokenized)
opt_detokenized = opt_sp.detokenize(tftext_tokenized)
self.assertAllEqual(tftext_detokenized, opt_detokenized)
def __init__(self, model, nbest_size=0, alpha=1.0):
"""Initializes the tokenizer.
Args:
model: Path to the SentencePiece model.
nbest_size: Number of candidates to sample from (disabled during inference).
alpha: Smoothing parameter for the sampling.
"""
super().__init__()
self._nbest_size = nbest_size
with tf.io.gfile.GFile(model, "rb") as model_file:
self._tokenizer = tft.SentencepieceTokenizer(
model=model_file.read(),
out_type=tf.string,
nbest_size=nbest_size,
alpha=alpha,
)
def __init__(self, vocab_file):
sentence_model_voca = vocab_file
sp_model = spm.SentencePieceProcessor()
sp_proto = tf.io.gfile.GFile(sentence_model_voca, "rb").read()
sp_model.LoadFromSerializedProto(sp_proto)
self.vocab_size = sp_model.GetPieceSize()
word_to_token = [sp_model.IdToPiece(i) for i in range(self.vocab_size)]
self.word_start_subtoken = np.array(
[sp_model.IdToPiece(i)[0] == "▁" for i in range(self.vocab_size)])
self.tf_tokenizer = tft.SentencepieceTokenizer(
model=tf.io.gfile.GFile(sentence_model_voca, "rb").read())
self.vocab = ko_bpe_vocab(word_to_token)
self.inv_vocab = {v: k for k, v in self.vocab.items()}
def _tokenize_example(example):
text, label = example["text"], example["label"]
tokenizer = tft.SentencepieceTokenizer(
model=tf.io.gfile.GFile(vocab_model_file, "rb").read())
if substitute_newline:
text = tf.strings.regex_replace(text, "\n", substitute_newline)
ids = tokenizer.tokenize(text)
ids = ids[:max_encoder_length - 2]
# Add [CLS] (65) and [SEP] (66) special tokens.
prefix = tf.constant([65])
suffix = tf.constant([66])
ids = tf.concat([prefix, ids, suffix], axis=0)
if isinstance(ids, tf.RaggedTensor):
ids = ids.to_tensor(0)
# tf.Example only supports tf.int64, but the TPU is better with tf.int32.
label = tf.cast(label, tf.int32)
return ids, label
def main(args: argparse.Namespace):
logger = get_logger("make-tfrecord")
input_files = glob.glob(args.dataset_paths)
logger.info(f"[+] Number of Dataset Files: {len(input_files)}")
# Load Config
logger.info(f"[+] Load Config From {args.data_config}")
config = DataConfig.from_yaml(args.data_config)
# Load Sentencepiece model
logger.info(f"[+] Load Tokenizer From {args.sp_model_path}")
with open(args.sp_model_path, "rb") as f:
tokenizer = text.SentencepieceTokenizer(f.read(),
add_bos=True,
add_eos=True)
serialize = tf.function(lambda audio, text: tf.io.serialize_tensor(
tf.stack([tf.io.serialize_tensor(audio),
tf.io.serialize_tensor(text)])))
logger.info("[+] Start Saving Dataset...")
for file_path in tqdm(input_files):
output_dir = args.output_dir if args.output_dir else os.path.dirname(
file_path)
file_name = os.path.basename(file_path)
output_path = os.path.join(
output_dir,
os.path.splitext(file_name)[0] + ".tfrecord")
# Write TFRecordFile
dataset = (get_dataset(
file_path, config.file_format, config.sample_rate, tokenizer).map(
config.audio_feature_fn,
num_parallel_calls=tf.data.experimental.AUTOTUNE).map(
serialize,
num_parallel_calls=tf.data.experimental.AUTOTUNE))
writer = tf.data.experimental.TFRecordWriter(output_path, "GZIP")
writer.write(dataset)
logger.info("[+] Done")
def encode(text: tf.Tensor, max_len: int, vocab_filename: str,
encoder_type: str):
"""EncodeOp."""
if encoder_type not in ["sentencepiece", "sentencepiece_newline"]:
raise ValueError("Unsupported encoder type: %s" % encoder_type)
sp_model = tf.gfile.GFile(vocab_filename, "rb").read()
tokenizer = tf_text.SentencepieceTokenizer(model=sp_model)
batch_size = text.shape[0]
if encoder_type == "sentencepiece_newline":
text = tf.strings.regex_replace(text, "\n", _NEWLINE_SYMBOL)
ids = tokenizer.tokenize(text)
eos = tf.ragged.constant([[1]] * batch_size)
ids = tf.concat([ids, eos], axis=1)
ids = ids.to_tensor(default_value=0)
ids = ids[:, :max_len]
pad = max_len - tf.shape(ids)[1]
ids = tf.pad(ids, [[0, 0], [0, pad]])
ids.set_shape([ids.shape[0], max_len])
ids = tf.where(ids > 1, ids + _SHIFT_RESERVED_TOKENS, ids)
ids = tf.cast(ids, tf.int64)
return ids
def __init__(self, params: cfg.TaskConfig, logging_dir=None, name=None):
super().__init__(params, logging_dir, name=name)
self._sentencepiece_model_path = params.sentencepiece_model_path
if params.sentencepiece_model_path:
self._sp_tokenizer = tftxt.SentencepieceTokenizer(
model=tf.io.gfile.GFile(params.sentencepiece_model_path, "rb").read(),
add_eos=True)
try:
empty_str_tokenized = self._sp_tokenizer.tokenize("").numpy()
except tf.errors.InternalError:
raise ValueError(
"EOS token not in tokenizer vocab."
"Please make sure the tokenizer generates a single token for an "
"empty string.")
self._eos_id = empty_str_tokenized.item()
self._vocab_size = self._sp_tokenizer.vocab_size().numpy()
else:
raise ValueError("Setencepiece model path not provided.")
if (params.validation_data.input_path or
params.validation_data.tfds_name) and self._logging_dir:
self._references, self._tf_record_input_path = write_test_record(
params.validation_data, self.logging_dir)
def benchmarkTokenizer(self):
sp_model = _GetSentencepieceModel()
test_text = [
"This week we celebrate the casts and creatives who have come together"
" to bring us our favorite.",
"More Stacks products demonstrated commitment to excellent support.",
"Test, test, test."
]
tftext_sp = tensorflow_text.SentencepieceTokenizer(sp_model)
opt_sp = sentencepiece_tokenizer.SentencepieceTokenizer(sp_model)
iter_number = 1000
start = time.time()
for _ in range(iter_number):
_ = opt_sp.tokenize(test_text)
self.report_benchmark(iters=iter_number,
wall_time=time.time() - start,
name="opt")
start = time.time()
for _ in range(iter_number):
_ = tftext_sp.tokenize(test_text)
self.report_benchmark(iters=iter_number,
wall_time=time.time() - start,
name="tf.text")
def __init__(self,
bert_layer,
max_len,
min_len=1,
CLS='[CLS]',
SEP='[SEP]',
PAD='[PAD]',
UNK='[UNK]'):
""" Initializes the layer
:param CLS Token that represents the start of a sentence
:param SEP Token that represents the end of a segment
:param PAD Token that represents padding
:param UNK Token that represents unknown tokens
:param bert_layer Keras layer that loaded from pretrained BERT
"""
super().__init__()
self._CLS = CLS
self._SEP = SEP
self._PAD = PAD
self._min_len = min_len
self._max_len = max_len
resolved_object = bert_layer.resolved_object
self.do_lower_case = resolved_object.do_lower_case.numpy()
if hasattr(resolved_object, "tokenizer_type"):
tokenizer_type_file = resolved_object.tokenizer_type.asset_path.numpy(
).decode("utf-8")
with tf.io.gfile.GFile(tokenizer_type_file, 'r') as f_handler:
self._tokenizer_type = f_handler.read().strip()
tokenizer_file = resolved_object.tokenizer_file.asset_path.numpy(
).decode("utf-8")
if self._tokenizer_type == SENTENCEPIECE:
with tf.io.gfile.GFile(tokenizer_file, 'rb') as f_handler:
sp_model = f_handler.read()
self._tokenizer = tf_text.SentencepieceTokenizer(
model=sp_model, out_type=tf.int32)
self.vocab_table = create_tf_vocab_from_sp_tokenizer(
self._tokenizer, num_oov_buckets=1)
else:
assert (self._tokenizer_type == SPACE)
_, self.vocab_table = read_tf_vocab(tokenizer_file, UNK)
else:
vocab_file = resolved_object.vocab_file.asset_path.numpy().decode(
"utf-8")
_, self.vocab_table = create_tf_vocab_from_wp_tokenizer(
vocab_file, num_oov_buckets=1)
self._tokenizer = tf_text.BertTokenizer(
self.vocab_table,
token_out_type=tf.int64,
lower_case=self.do_lower_case,
unknown_token=UNK)
self._tokenizer_type = WORDPIECE
self._pad_id = self.vocab_table.lookup(tf.constant(PAD)) if PAD else -1
self._cls_id = self.vocab_table.lookup(tf.constant(CLS)) if CLS else -1
self._sep_id = self.vocab_table.lookup(tf.constant(SEP)) if SEP else -1
if self._tokenizer_type == SENTENCEPIECE:
self._pad_id = tf.cast(self._pad_id, tf.int32)
self._cls_id = tf.cast(self._cls_id, tf.int32)
self._sep_id = tf.cast(self._sep_id, tf.int32)
def _create_tokenizer(self):
return text.SentencepieceTokenizer(model=self._model_serialized_proto,
out_type=tf.int32,
nbest_size=self._nbest_size,
alpha=self._alpha)
def tf_tokenizer(self):
"""Instantiate and return a TF tokenizer."""
return tf_text.SentencepieceTokenizer(model=self.sp_model)
def load_sentencepiece_model(model_proto):
proto = tf.io.gfile.GFile(model_proto, 'rb').read()
return tf_text.SentencepieceTokenizer(model=proto)
def __init__(self, vocab_file):
super().__init__()
serialized_proto = tf.compat.v1.gfile.GFile(vocab_file, "rb").read()
self.tokenizer = tf_text.SentencepieceTokenizer(
model=serialized_proto, add_bos=True, add_eos=True)
# fmt: on
if __name__ == "__main__":
args = parser.parse_args()
strategy = get_device_strategy(args.device)
logger = get_logger()
if args.mixed_precision:
policy = tf.keras.mixed_precision.experimental.Policy("mixed_float16")
tf.keras.mixed_precision.experimental.set_policy(policy)
logger.info("Use Mixed Precision FP16")
# Construct Dataset
with tf.io.gfile.GFile(args.sp_model_path, "rb") as f:
tokenizer = text.SentencepieceTokenizer(f.read(), add_bos=True, add_eos=True)
dataset_files = tf.io.gfile.glob(args.dataset_path)
if not dataset_files:
logger.error("[Error] Dataset path is invalid!")
sys.exit(1)
if args.auto_encoding:
scatter = lambda tokens: (tokens, tokens)
dataset = (
tf.data.TextLineDataset(dataset_files, num_parallel_reads=tf.data.experimental.AUTOTUNE)
.map(tokenizer.tokenize)
.map(scatter)
)
else:
tokenize = lambda inputs, outputs: ((tokenizer.tokenize(inputs), tokenizer.tokenize(outputs)))
dataset = tf.data.experimental.CsvDataset(
def __call__(self, x):
# Constrained sequence
cs_scores = np.array([[10.0, 12.0, 6.0, 4.0], [13.0, 12.0, 11.0,
10.0]])
cs_input = np.array([cs_scores, cs_scores, cs_scores],
dtype=np.float32)
cs_transition_weights = np.array(
[[-1.0, 1.0, -2.0, 2.0, 0.0], [3.0, -3.0, 4.0, -4.0, 0.0],
[5.0, 1.0, 10.0, 1.0, 1.0], [-7.0, 7.0, -8.0, 8.0, 0.0],
[0.0, 1.0, 2.0, 3.0, 0.0]],
dtype=np.float32)
cs_allowed_transitions = np.array([[True, True, True, True, True],
[True, True, True, True, True],
[True, False, True, False, False],
[True, True, True, True, True],
[True, False, True, True, True]])
constrained_sequence = text.viterbi_constrained_sequence(
cs_input, [2, 2, 2],
allowed_transitions=cs_allowed_transitions,
transition_weights=cs_transition_weights,
use_log_space=True,
use_start_and_end_states=True)
# Max Spanning Tree
mst_num_nodes = tf.constant([4, 3], tf.int32)
mst_scores = tf.constant(
[[[0, 0, 0, 0], [1, 0, 0, 0], [1, 2, 0, 0], [1, 2, 3, 4]],
[[4, 3, 2, 9], [0, 0, 2, 9], [0, 0, 0, 9], [9, 9, 9, 9]]],
tf.int32) # pyformat: disable
(max_spanning_tree,
_) = text.max_spanning_tree(mst_num_nodes, mst_scores)
# Normalize
normalized = text.case_fold_utf8(['A String'])
normalized = text.normalize_utf8(normalized)
# Regex split
regex_split = text.regex_split(input=['Yo dawg!'],
delim_regex_pattern=r'\s')
# Rouge-L
rl_hypotheses = tf.ragged.constant(
[['captain', 'of', 'the', 'delta', 'flight'],
['the', '1990', 'transcript']])
rl_references = tf.ragged.constant(
[['delta', 'air', 'lines', 'flight'],
['this', 'concludes', 'the', 'transcript']])
(rouge_l, _, _) = text.metrics.rouge_l(rl_hypotheses, rl_references)
# Sentence breaking version 1 (token dependent)
sb_token_word = [['Welcome', 'to', 'the', 'U.S.', '!', 'Harry'],
['Wu', 'Tang', 'Clan', ';', 'ain\'t', 'nothing']]
sb_token_properties = [[0, 0, 0, 256, 0, 0], [0, 0, 0, 0, 0, 0]]
sb_token_starts = []
sb_token_ends = []
for sentence in sb_token_word:
sentence_string = ''
sentence_start = []
sentence_end = []
for word in sentence:
sentence_start.append(len(sentence_string))
sentence_string = sentence_string.join([word, ' '])
sentence_end.append(len(sentence_string))
sb_token_starts.append(sentence_start)
sb_token_ends.append(sentence_end)
sb_token_starts = tf.constant(sb_token_starts, dtype=tf.int64)
sb_token_ends = tf.constant(sb_token_ends, dtype=tf.int64)
sb_token_properties = tf.ragged.constant(sb_token_properties,
dtype=tf.int64)
(sentence_breaking, _, _,
_) = text.sentence_fragments(sb_token_word, sb_token_starts,
sb_token_ends, sb_token_properties)
# Sentence breaking version 2 (StateBasedSentenceBreaker)
sbv2_text_input = [['Welcome to the U.S.! Harry'],
['Wu Tang Clan; ain\'t nothing']]
sentence_breaker_v2 = text.StateBasedSentenceBreaker()
sbv2_fragment_text, _, _ = (
sentence_breaker_v2.break_sentences_with_offsets(sbv2_text_input))
# Sentencepiece tokenizer
sp_model_file = (
'third_party/tensorflow_text/python/ops/test_data/test_oss_model.model'
)
sp_model = open(sp_model_file, 'rb').read()
sp_tokenizer = text.SentencepieceTokenizer(sp_model)
sentencepiece = sp_tokenizer.tokenize(['A sentence of things.'])
sentencepiece = sp_tokenizer.detokenize(sentencepiece)
(sentencepiece, _,
_) = sp_tokenizer.tokenize_with_offsets(sentencepiece)
sentencepiece_size = sp_tokenizer.vocab_size()
sentencepiece_id = sp_tokenizer.id_to_string(1)
# Split merge tokenizer
sm_tokenizer = text.SplitMergeTokenizer()
split_merge = sm_tokenizer.tokenize(b'IloveFlume!',
[0, 0, 1, 1, 1, 0, 1, 1, 1, 1, 0])
# Split merge from logits tokenizer
smfl_tokenizer = text.SplitMergeFromLogitsTokenizer()
split_merge_from_logits = smfl_tokenizer.tokenize(
b'IloveFlume!',
# One pair of logits for each Unicode character from the text. Each
# pair indicates a "split" action if the first component is greater than
# the second one, and a "merge" otherwise.
[
[2.7, -0.3], # I: split
[4.1, 0.82], # l: split
[-2.3, 4.3], # o: merge
[3.1, 12.2], # v: merge
[-3.0, 4.7], # e: merge
[2.7, -0.7], # F: split
[0.7, 15.0], # l: merge
[1.6, 23.0], # u: merge
[2.1, 11.0], # m: merge
[0.0, 20.0], # e: merge
[18.0, 0.7], # !: split
])
# Confirm TF unicode_script op that requires ICU works
tf_unicode_script = tf.strings.unicode_script(
[ord('a'), 0x0411, 0x82b8, ord(',')])
# Unicode script tokenizer
us_tokenizer = text.UnicodeScriptTokenizer()
unicode_script = us_tokenizer.tokenize(['a string'])
# Whitespace tokenizer
ws_tokenizer = text.WhitespaceTokenizer()
whitespace = ws_tokenizer.tokenize(['a string'])
# Wordpiece tokenizer
wp_initializer = tf.lookup.KeyValueTensorInitializer(
['i'], [1], key_dtype=tf.string, value_dtype=tf.int64)
self.wp_vocab_table = tf.lookup.StaticHashTable(wp_initializer,
default_value=-1)
wp_tokenizer = text.WordpieceTokenizer(self.wp_vocab_table)
wordpiece = wp_tokenizer.tokenize(['i am'])
# Wordshape
wordshapes = text.wordshape([u'a-b', u'a\u2010b'.encode('utf-8')],
text.WordShape.HAS_PUNCTUATION_DASH)
# Assertion method
def assert_check(tensor):
return tf.assert_equal(tensor, tf.identity(tensor))
# Assertions
constrained_sequence_assert = assert_check(
constrained_sequence.to_tensor())
max_spanning_tree_assert = assert_check(max_spanning_tree)
normalized_assert = assert_check(normalized)
regex_split_assert = assert_check(regex_split.to_tensor())
rouge_l_assert = assert_check(rouge_l)
sentence_breaking_assert = assert_check(sentence_breaking.to_tensor())
sentence_breaking_v2_assert = assert_check(
sbv2_fragment_text.to_tensor())
sentencepiece_assert = assert_check(sentencepiece.to_tensor())
sentencepiece_id_assert = assert_check(sentencepiece_id)
sentencepiece_size_assert = assert_check(sentencepiece_size)
split_merge_assert = assert_check(split_merge)
split_merge_from_logits_assert = assert_check(split_merge_from_logits)
tf_unicode_script_assert = assert_check(tf_unicode_script)
unicode_script_assert = assert_check(unicode_script.to_tensor())
whitespace_assert = assert_check(whitespace.to_tensor())
wordpiece_assert = assert_check(wordpiece.to_tensor())
wordshapes_assert = assert_check(wordshapes)
with tf.control_dependencies([
constrained_sequence_assert, max_spanning_tree_assert,
normalized_assert, regex_split_assert, rouge_l_assert,
sentence_breaking_assert, sentence_breaking_v2_assert,
sentencepiece_assert, sentencepiece_id_assert,
sentencepiece_size_assert, split_merge_assert,
split_merge_from_logits_assert, tf_unicode_script_assert,
unicode_script_assert, whitespace_assert, wordpiece_assert,
wordshapes_assert
]):
y = tf.add(x, [1])
return {'y': y}
