def __init__(self, checkpoint=None, predict_fn=None):
"""Initializes the BLEURT model.
Args:
checkpoint: BLEURT checkpoint. Will default to BLEURT-tiny if None.
predict_fn: (optional) prediction function, overrides chkpt_dir. Mostly
used for testing.
Returns:
A BLEURT scorer export.
"""
if not checkpoint:
logging.info("No checkpoint specified, defaulting to BLEURT-tiny.")
checkpoint = _get_default_checkpoint()
logging.info("Reading checkpoint {}.".format(checkpoint))
config = checkpoint_lib.read_bleurt_config(checkpoint)
max_seq_length = config["max_seq_length"]
vocab_file = config["vocab_file"]
do_lower_case = config["do_lower_case"]
logging.info("Creating BLEURT scorer.")
self.tokenizer = tokenization.FullTokenizer(
vocab_file=vocab_file, do_lower_case=do_lower_case)
self.max_seq_length = max_seq_length
if predict_fn:
self.predict_fn = predict_fn
logging.info("BLEURT initialized.")
return
logging.info("Loading model...")
self.chkpt_dir = checkpoint
self.predict_fn = _make_predict_fn_from_checkpoint(checkpoint)
logging.info("BLEURT initialized.")
def encode_and_serialize(input_file, output_file, vocab_file, do_lower_case,
max_seq_length):
"""Encodes and serializes a set of ratings in JSON format."""
assert tf.io.gfile.exists(input_file), "Could not find file."
logging.info("Reading data...")
with tf.io.gfile.GFile(input_file, "r") as f:
examples_df = pd.read_json(f, lines=True)
for col in ["reference", "candidate", "score"]:
assert col in examples_df.columns, \
"field {} not found in input file!".format(col)
n_records = len(examples_df)
logging.info("Read {} examples.".format(n_records))
logging.info("Encoding and writing TFRecord file...")
tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=do_lower_case)
with tf.python_io.TFRecordWriter(output_file) as writer:
iterator_id, iterator_cycle = 0, max(int(n_records / 10), 1)
for record in examples_df.itertuples(index=False):
iterator_id += 1
if iterator_id % iterator_cycle == 0:
logging.info("Writing example %d of %d", iterator_id,
n_records)
tf_example = serialize_example(record.reference,
record.candidate,
tokenizer,
max_seq_length,
score=record.score)
writer.write(tf_example)
logging.info("Done writing {} tf examples.".format(n_records))
def bleurt_ops(references, candidates):
"""Builds computation graph for BLEURT.
Args:
references: <tf.string>[...] Tensor that contains reference sentences.
candidates: <tf.string>[...] Tensor that contains candidate sentences.
Returns:
A <tf.float>[...] Tensor that contains BLEURT scores.
"""
logging.info("Creating BLEURT TF Ops...")
logging.info("Reading info from checkpoint {}".format(checkpoint))
config = checkpoint_lib.read_bleurt_config(checkpoint)
max_seq_length = config["max_seq_length"]
vocab_file = config["vocab_file"]
do_lower_case = config["do_lower_case"]
logging.info("Creating tokenizer...")
tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=do_lower_case)
logging.info("Tokenizer created")
logging.info("Creating BLEURT Preprocessing Ops...")
bleurt_preprocessing_ops = create_bleurt_preprocessing_ops(
tokenizer, max_seq_length)
logging.info("Preprocessing Ops created.")
logging.info("Loading checkpoint...")
if not bleurt_model_fn:
imported = tf.saved_model.load_v2(checkpoint)
bleurt_model_ops = imported.signatures["serving_default"]
else:
bleurt_model_ops = bleurt_model_fn
logging.info("BLEURT Checkpoint loaded")
input_ids, input_mask, segment_ids = bleurt_preprocessing_ops(
references, candidates)
out = bleurt_model_ops(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids)
logging.info("BLEURT TF Ops created.")
return out
