Skip to content

GitLab

Commit 8163baab authored by Tim Rault's avatar Tim Rault
Browse files

Convert indentation from 2 spaces to 4 spaces

parent 555b7d66
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 3917 additions and 3918 deletions
create_pretraining_data.py
View file @ 8163baab
This diff is collapsed.
extract_features.py
View file @ 8163baab
This diff is collapsed.
modeling.py
View file @ 8163baab
This diff is collapsed.
modeling_test.py
View file @ 8163baab
......@@ -27,250 +27,249 @@ import tensorflow as tf
class BertModelTest(tf.test.TestCase):
class BertModelTester(object):
def __init__(self,
parent,
batch_size=13,
seq_length=7,
is_training=True,
use_input_mask=True,
use_token_type_ids=True,
vocab_size=99,
hidden_size=32,
num_hidden_layers=5,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
initializer_range=0.02,
scope=None):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_input_mask = use_input_mask
self.use_token_type_ids = use_token_type_ids
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.initializer_range = initializer_range
self.scope = scope
def create_model(self):
input_ids = BertModelTest.ids_tensor([self.batch_size, self.seq_length],
self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = BertModelTest.ids_tensor(
[self.batch_size, self.seq_length], vocab_size=2)
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = BertModelTest.ids_tensor(
[self.batch_size, self.seq_length], self.type_vocab_size)
config = modeling.BertConfig(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range)
model = modeling.BertModel(
config=config,
is_training=self.is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=token_type_ids,
scope=self.scope)
outputs = {
"embedding_output": model.get_embedding_output(),
"sequence_output": model.get_sequence_output(),
"pooled_output": model.get_pooled_output(),
"all_encoder_layers": model.get_all_encoder_layers(),
}
return outputs
def check_output(self, result):
self.parent.assertAllEqual(
result["embedding_output"].shape,
[self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertAllEqual(
result["sequence_output"].shape,
[self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertAllEqual(result["pooled_output"].shape,
[self.batch_size, self.hidden_size])
def test_default(self):
self.run_tester(BertModelTest.BertModelTester(self))
def test_config_to_json_string(self):
config = modeling.BertConfig(vocab_size=99, hidden_size=37)
obj = json.loads(config.to_json_string())
self.assertEqual(obj["vocab_size"], 99)
self.assertEqual(obj["hidden_size"], 37)
def run_tester(self, tester):
with self.test_session() as sess:
ops = tester.create_model()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
output_result = sess.run(ops)
tester.check_output(output_result)
self.assert_all_tensors_reachable(sess, [init_op, ops])
@classmethod
def ids_tensor(cls, shape, vocab_size, rng=None, name=None):
"""Creates a random int32 tensor of the shape within the vocab size."""
if rng is None:
rng = random.Random()
total_dims = 1
for dim in shape:
total_dims *= dim
values = []
for _ in range(total_dims):
values.append(rng.randint(0, vocab_size - 1))
return tf.constant(value=values, dtype=tf.int32, shape=shape, name=name)
def assert_all_tensors_reachable(self, sess, outputs):
"""Checks that all the tensors in the graph are reachable from outputs."""
graph = sess.graph
ignore_strings = [
"^.*/dilation_rate$",
"^.*/Tensordot/concat$",
"^.*/Tensordot/concat/axis$",
"^testing/.*$",
]
ignore_regexes = [re.compile(x) for x in ignore_strings]
unreachable = self.get_unreachable_ops(graph, outputs)
filtered_unreachable = []
for x in unreachable:
do_ignore = False
for r in ignore_regexes:
m = r.match(x.name)
if m is not None:
do_ignore = True
if do_ignore:
continue
filtered_unreachable.append(x)
unreachable = filtered_unreachable
self.assertEqual(
len(unreachable), 0, "The following ops are unreachable: %s" %
(" ".join([x.name for x in unreachable])))
@classmethod
def get_unreachable_ops(cls, graph, outputs):
"""Finds all of the tensors in graph that are unreachable from outputs."""
outputs = cls.flatten_recursive(outputs)
output_to_op = collections.defaultdict(list)
op_to_all = collections.defaultdict(list)
assign_out_to_in = collections.defaultdict(list)
for op in graph.get_operations():
for x in op.inputs:
op_to_all[op.name].append(x.name)
for y in op.outputs:
output_to_op[y.name].append(op.name)
op_to_all[op.name].append(y.name)
if str(op.type) == "Assign":
for y in op.outputs:
for x in op.inputs:
assign_out_to_in[y.name].append(x.name)
assign_groups = collections.defaultdict(list)
for out_name in assign_out_to_in.keys():
name_group = assign_out_to_in[out_name]
for n1 in name_group:
assign_groups[n1].append(out_name)
for n2 in name_group:
if n1 != n2:
assign_groups[n1].append(n2)
seen_tensors = {}
stack = [x.name for x in outputs]
while stack:
name = stack.pop()
if name in seen_tensors:
continue
seen_tensors[name] = True
if name in output_to_op:
for op_name in output_to_op[name]:
if op_name in op_to_all:
for input_name in op_to_all[op_name]:
if input_name not in stack:
stack.append(input_name)
expanded_names = []
if name in assign_groups:
for assign_name in assign_groups[name]:
expanded_names.append(assign_name)
for expanded_name in expanded_names:
if expanded_name not in stack:
stack.append(expanded_name)
unreachable_ops = []
for op in graph.get_operations():
is_unreachable = False
all_names = [x.name for x in op.inputs] + [x.name for x in op.outputs]
for name in all_names:
if name not in seen_tensors:
is_unreachable = True
if is_unreachable:
unreachable_ops.append(op)
return unreachable_ops
@classmethod
def flatten_recursive(cls, item):
"""Flattens (potentially nested) a tuple/dictionary/list to a list."""
output = []
if isinstance(item, list):
output.extend(item)
elif isinstance(item, tuple):
output.extend(list(item))
elif isinstance(item, dict):
for (_, v) in six.iteritems(item):
output.append(v)
else:
return [item]
flat_output = []
for x in output:
flat_output.extend(cls.flatten_recursive(x))
return flat_output
class BertModelTester(object):
def __init__(self,
parent,
batch_size=13,
seq_length=7,
is_training=True,
use_input_mask=True,
use_token_type_ids=True,
vocab_size=99,
hidden_size=32,
num_hidden_layers=5,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
initializer_range=0.02,
scope=None):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_input_mask = use_input_mask
self.use_token_type_ids = use_token_type_ids
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.initializer_range = initializer_range
self.scope = scope
def create_model(self):
input_ids = BertModelTest.ids_tensor([self.batch_size, self.seq_length],
self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = BertModelTest.ids_tensor(
[self.batch_size, self.seq_length], vocab_size=2)
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = BertModelTest.ids_tensor(
[self.batch_size, self.seq_length], self.type_vocab_size)
config = modeling.BertConfig(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range)
model = modeling.BertModel(
config=config,
is_training=self.is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=token_type_ids,
scope=self.scope)
outputs = {
"embedding_output": model.get_embedding_output(),
"sequence_output": model.get_sequence_output(),
"pooled_output": model.get_pooled_output(),
"all_encoder_layers": model.get_all_encoder_layers(),
}
return outputs
def check_output(self, result):
self.parent.assertAllEqual(
result["embedding_output"].shape,
[self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertAllEqual(
result["sequence_output"].shape,
[self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertAllEqual(result["pooled_output"].shape,
[self.batch_size, self.hidden_size])
def test_default(self):
self.run_tester(BertModelTest.BertModelTester(self))
def test_config_to_json_string(self):
config = modeling.BertConfig(vocab_size=99, hidden_size=37)
obj = json.loads(config.to_json_string())
self.assertEqual(obj["vocab_size"], 99)
self.assertEqual(obj["hidden_size"], 37)
def run_tester(self, tester):
with self.test_session() as sess:
ops = tester.create_model()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
output_result = sess.run(ops)
tester.check_output(output_result)
self.assert_all_tensors_reachable(sess, [init_op, ops])
@classmethod
def ids_tensor(cls, shape, vocab_size, rng=None, name=None):
"""Creates a random int32 tensor of the shape within the vocab size."""
if rng is None:
rng = random.Random()
total_dims = 1
for dim in shape:
total_dims *= dim
values = []
for _ in range(total_dims):
values.append(rng.randint(0, vocab_size - 1))
return tf.constant(value=values, dtype=tf.int32, shape=shape, name=name)
def assert_all_tensors_reachable(self, sess, outputs):
"""Checks that all the tensors in the graph are reachable from outputs."""
graph = sess.graph
ignore_strings = [
"^.*/dilation_rate$",
"^.*/Tensordot/concat$",
"^.*/Tensordot/concat/axis$",
"^testing/.*$",
]
ignore_regexes = [re.compile(x) for x in ignore_strings]
unreachable = self.get_unreachable_ops(graph, outputs)
filtered_unreachable = []
for x in unreachable:
do_ignore = False
for r in ignore_regexes:
m = r.match(x.name)
if m is not None:
do_ignore = True
if do_ignore:
continue
filtered_unreachable.append(x)
unreachable = filtered_unreachable
self.assertEqual(
len(unreachable), 0, "The following ops are unreachable: %s" %
(" ".join([x.name for x in unreachable])))
@classmethod
def get_unreachable_ops(cls, graph, outputs):
"""Finds all of the tensors in graph that are unreachable from outputs."""
outputs = cls.flatten_recursive(outputs)
output_to_op = collections.defaultdict(list)
op_to_all = collections.defaultdict(list)
assign_out_to_in = collections.defaultdict(list)
for op in graph.get_operations():
for x in op.inputs:
op_to_all[op.name].append(x.name)
for y in op.outputs:
output_to_op[y.name].append(op.name)
op_to_all[op.name].append(y.name)
if str(op.type) == "Assign":
for y in op.outputs:
for x in op.inputs:
assign_out_to_in[y.name].append(x.name)
assign_groups = collections.defaultdict(list)
for out_name in assign_out_to_in.keys():
name_group = assign_out_to_in[out_name]
for n1 in name_group:
assign_groups[n1].append(out_name)
for n2 in name_group:
if n1 != n2:
assign_groups[n1].append(n2)
seen_tensors = {}
stack = [x.name for x in outputs]
while stack:
name = stack.pop()
if name in seen_tensors:
continue
seen_tensors[name] = True
if name in output_to_op:
for op_name in output_to_op[name]:
if op_name in op_to_all:
for input_name in op_to_all[op_name]:
if input_name not in stack:
stack.append(input_name)
expanded_names = []
if name in assign_groups:
for assign_name in assign_groups[name]:
expanded_names.append(assign_name)
for expanded_name in expanded_names:
if expanded_name not in stack:
stack.append(expanded_name)
unreachable_ops = []
for op in graph.get_operations():
is_unreachable = False
all_names = [x.name for x in op.inputs] + [x.name for x in op.outputs]
for name in all_names:
if name not in seen_tensors:
is_unreachable = True
if is_unreachable:
unreachable_ops.append(op)
return unreachable_ops
@classmethod
def flatten_recursive(cls, item):
"""Flattens (potentially nested) a tuple/dictionary/list to a list."""
output = []
if isinstance(item, list):
output.extend(item)
elif isinstance(item, tuple):
output.extend(list(item))
elif isinstance(item, dict):
for (_, v) in six.iteritems(item):
output.append(v)
else:
return [item]
flat_output = []
for x in output:
flat_output.extend(cls.flatten_recursive(x))
return flat_output
if __name__ == "__main__":
tf.test.main()
tf.test.main()
optimization.py
View file @ 8163baab
......@@ -23,149 +23,149 @@ import tensorflow as tf
def create_optimizer(loss, init_lr, num_train_steps, num_warmup_steps, use_tpu):
"""Creates an optimizer training op."""
global_step = tf.train.get_or_create_global_step()
"""Creates an optimizer training op."""
global_step = tf.train.get_or_create_global_step()
learning_rate = tf.constant(value=init_lr, shape=[], dtype=tf.float32)
learning_rate = tf.constant(value=init_lr, shape=[], dtype=tf.float32)
# Implements linear decay of the learning rate.
learning_rate = tf.train.polynomial_decay(
learning_rate,
global_step,
num_train_steps,
end_learning_rate=0.0,
power=1.0,
cycle=False)
# Implements linear decay of the learning rate.
learning_rate = tf.train.polynomial_decay(
learning_rate,
global_step,
num_train_steps,
end_learning_rate=0.0,
power=1.0,
cycle=False)
# Implements linear warmup. I.e., if global_step < num_warmup_steps, the
# learning rate will be `global_step/num_warmup_steps * init_lr`.
if num_warmup_steps:
global_steps_int = tf.cast(global_step, tf.int32)
warmup_steps_int = tf.constant(num_warmup_steps, dtype=tf.int32)
# Implements linear warmup. I.e., if global_step < num_warmup_steps, the
# learning rate will be `global_step/num_warmup_steps * init_lr`.
if num_warmup_steps:
global_steps_int = tf.cast(global_step, tf.int32)
warmup_steps_int = tf.constant(num_warmup_steps, dtype=tf.int32)
global_steps_float = tf.cast(global_steps_int, tf.float32)
warmup_steps_float = tf.cast(warmup_steps_int, tf.float32)
global_steps_float = tf.cast(global_steps_int, tf.float32)
warmup_steps_float = tf.cast(warmup_steps_int, tf.float32)
warmup_percent_done = global_steps_float / warmup_steps_float
warmup_learning_rate = init_lr * warmup_percent_done
warmup_percent_done = global_steps_float / warmup_steps_float
warmup_learning_rate = init_lr * warmup_percent_done
is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32)
learning_rate = (
(1.0 - is_warmup) * learning_rate + is_warmup * warmup_learning_rate)
is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32)
learning_rate = (
(1.0 - is_warmup) * learning_rate + is_warmup * warmup_learning_rate)
# It is recommended that you use this optimizer for fine tuning, since this
# is how the model was trained (note that the Adam m/v variables are NOT
# loaded from init_checkpoint.)
optimizer = AdamWeightDecayOptimizer(
learning_rate=learning_rate,
weight_decay_rate=0.01,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=["LayerNorm", "layer_norm", "bias"])
# It is recommended that you use this optimizer for fine tuning, since this
# is how the model was trained (note that the Adam m/v variables are NOT
# loaded from init_checkpoint.)
optimizer = AdamWeightDecayOptimizer(
learning_rate=learning_rate,
weight_decay_rate=0.01,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=["LayerNorm", "layer_norm", "bias"])
if use_tpu:
optimizer = tf.contrib.tpu.CrossShardOptimizer(optimizer)
if use_tpu:
optimizer = tf.contrib.tpu.CrossShardOptimizer(optimizer)
tvars = tf.trainable_variables()
grads = tf.gradients(loss, tvars)
tvars = tf.trainable_variables()
grads = tf.gradients(loss, tvars)
# This is how the model was pre-trained.
(grads, _) = tf.clip_by_global_norm(grads, clip_norm=1.0)
# This is how the model was pre-trained.
(grads, _) = tf.clip_by_global_norm(grads, clip_norm=1.0)
train_op = optimizer.apply_gradients(
zip(grads, tvars), global_step=global_step)
train_op = optimizer.apply_gradients(
zip(grads, tvars), global_step=global_step)
new_global_step = global_step + 1
train_op = tf.group(train_op, [global_step.assign(new_global_step)])
return train_op
new_global_step = global_step + 1
train_op = tf.group(train_op, [global_step.assign(new_global_step)])
return train_op
class AdamWeightDecayOptimizer(tf.train.Optimizer):
"""A basic Adam optimizer that includes "correct" L2 weight decay."""
def __init__(self,
learning_rate,
weight_decay_rate=0.0,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=None,
name="AdamWeightDecayOptimizer"):
"""Constructs a AdamWeightDecayOptimizer."""
super(AdamWeightDecayOptimizer, self).__init__(False, name)
self.learning_rate = learning_rate
self.weight_decay_rate = weight_decay_rate
self.beta_1 = beta_1
self.beta_2 = beta_2
self.epsilon = epsilon
self.exclude_from_weight_decay = exclude_from_weight_decay
def apply_gradients(self, grads_and_vars, global_step=None, name=None):
"""See base class."""
assignments = []
for (grad, param) in grads_and_vars:
if grad is None or param is None:
continue
param_name = self._get_variable_name(param.name)
m = tf.get_variable(
name=param_name + "/adam_m",
shape=param.shape.as_list(),
dtype=tf.float32,
trainable=False,
initializer=tf.zeros_initializer())
v = tf.get_variable(
name=param_name + "/adam_v",
shape=param.shape.as_list(),
dtype=tf.float32,
trainable=False,
initializer=tf.zeros_initializer())
# Standard Adam update.
next_m = (
tf.multiply(self.beta_1, m) + tf.multiply(1.0 - self.beta_1, grad))
next_v = (
tf.multiply(self.beta_2, v) + tf.multiply(1.0 - self.beta_2,
tf.square(grad)))
update = next_m / (tf.sqrt(next_v) + self.epsilon)
# Just adding the square of the weights to the loss function is *not*
# the correct way of using L2 regularization/weight decay with Adam,
# since that will interact with the m and v parameters in strange ways.
#
# Instead we want ot decay the weights in a manner that doesn't interact
# with the m/v parameters. This is equivalent to adding the square
# of the weights to the loss with plain (non-momentum) SGD.
if self._do_use_weight_decay(param_name):
update += self.weight_decay_rate * param
update_with_lr = self.learning_rate * update
next_param = param - update_with_lr
assignments.extend(
[param.assign(next_param),
m.assign(next_m),
v.assign(next_v)])
return tf.group(*assignments, name=name)
def _do_use_weight_decay(self, param_name):
"""Whether to use L2 weight decay for `param_name`."""
if not self.weight_decay_rate:
return False
if self.exclude_from_weight_decay:
for r in self.exclude_from_weight_decay:
if re.search(r, param_name) is not None:
return False
return True
def _get_variable_name(self, param_name):
"""Get the variable name from the tensor name."""
m = re.match("^(.*):\\d+$", param_name)
if m is not None:
param_name = m.group(1)
return param_name
"""A basic Adam optimizer that includes "correct" L2 weight decay."""
def __init__(self,
learning_rate,
weight_decay_rate=0.0,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=None,
name="AdamWeightDecayOptimizer"):
"""Constructs a AdamWeightDecayOptimizer."""
super(AdamWeightDecayOptimizer, self).__init__(False, name)
self.learning_rate = learning_rate
self.weight_decay_rate = weight_decay_rate
self.beta_1 = beta_1
self.beta_2 = beta_2
self.epsilon = epsilon
self.exclude_from_weight_decay = exclude_from_weight_decay
def apply_gradients(self, grads_and_vars, global_step=None, name=None):
"""See base class."""
assignments = []
for (grad, param) in grads_and_vars:
if grad is None or param is None:
continue
param_name = self._get_variable_name(param.name)
m = tf.get_variable(
name=param_name + "/adam_m",
shape=param.shape.as_list(),
dtype=tf.float32,
trainable=False,
initializer=tf.zeros_initializer())
v = tf.get_variable(
name=param_name + "/adam_v",
shape=param.shape.as_list(),
dtype=tf.float32,
trainable=False,
initializer=tf.zeros_initializer())
# Standard Adam update.
next_m = (
tf.multiply(self.beta_1, m) + tf.multiply(1.0 - self.beta_1, grad))
next_v = (
tf.multiply(self.beta_2, v) + tf.multiply(1.0 - self.beta_2,
tf.square(grad)))
update = next_m / (tf.sqrt(next_v) + self.epsilon)
# Just adding the square of the weights to the loss function is *not*
# the correct way of using L2 regularization/weight decay with Adam,
# since that will interact with the m and v parameters in strange ways.
#
# Instead we want ot decay the weights in a manner that doesn't interact
# with the m/v parameters. This is equivalent to adding the square
# of the weights to the loss with plain (non-momentum) SGD.
if self._do_use_weight_decay(param_name):
update += self.weight_decay_rate * param
update_with_lr = self.learning_rate * update
next_param = param - update_with_lr
assignments.extend(
[param.assign(next_param),
m.assign(next_m),
v.assign(next_v)])
return tf.group(*assignments, name=name)
def _do_use_weight_decay(self, param_name):
"""Whether to use L2 weight decay for `param_name`."""
if not self.weight_decay_rate:
return False
if self.exclude_from_weight_decay:
for r in self.exclude_from_weight_decay:
if re.search(r, param_name) is not None:
return False
return True
def _get_variable_name(self, param_name):
"""Get the variable name from the tensor name."""
m = re.match("^(.*):\\d+$", param_name)
if m is not None:
param_name = m.group(1)
return param_name
optimization_test.py
View file @ 8163baab
......@@ -22,27 +22,27 @@ import tensorflow as tf
class OptimizationTest(tf.test.TestCase):
def test_adam(self):
with self.test_session() as sess:
w = tf.get_variable(
"w",
shape=[3],
initializer=tf.constant_initializer([0.1, -0.2, -0.1]))
x = tf.constant([0.4, 0.2, -0.5])
loss = tf.reduce_mean(tf.square(x - w))
tvars = tf.trainable_variables()
grads = tf.gradients(loss, tvars)
global_step = tf.train.get_or_create_global_step()
optimizer = optimization.AdamWeightDecayOptimizer(learning_rate=0.2)
train_op = optimizer.apply_gradients(zip(grads, tvars), global_step)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
for _ in range(100):
sess.run(train_op)
w_np = sess.run(w)
self.assertAllClose(w_np.flat, [0.4, 0.2, -0.5], rtol=1e-2, atol=1e-2)
def test_adam(self):
with self.test_session() as sess:
w = tf.get_variable(
"w",
shape=[3],
initializer=tf.constant_initializer([0.1, -0.2, -0.1]))
x = tf.constant([0.4, 0.2, -0.5])
loss = tf.reduce_mean(tf.square(x - w))
tvars = tf.trainable_variables()
grads = tf.gradients(loss, tvars)
global_step = tf.train.get_or_create_global_step()
optimizer = optimization.AdamWeightDecayOptimizer(learning_rate=0.2)
train_op = optimizer.apply_gradients(zip(grads, tvars), global_step)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
for _ in range(100):
sess.run(train_op)
w_np = sess.run(w)
self.assertAllClose(w_np.flat, [0.4, 0.2, -0.5], rtol=1e-2, atol=1e-2)
if __name__ == "__main__":
tf.test.main()
tf.test.main()
run_classifier.py
View file @ 8163baab
This diff is collapsed.
run_pretraining.py
View file @ 8163baab
This diff is collapsed.
run_squad.py
View file @ 8163baab
This diff is collapsed.
tokenization.py
View file @ 8163baab
This diff is collapsed.
tokenization_test.py
View file @ 8163baab
......@@ -25,101 +25,101 @@ import tensorflow as tf
class TokenizationTest(tf.test.TestCase):
def test_full_tokenizer(self):
vocab_tokens = [
"[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn",
"##ing", ","
]
with tempfile.NamedTemporaryFile(delete=False) as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
def test_full_tokenizer(self):
vocab_tokens = [
"[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn",
"##ing", ","
]
with tempfile.NamedTemporaryFile(delete=False) as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
vocab_file = vocab_writer.name
vocab_file = vocab_writer.name
tokenizer = tokenization.FullTokenizer(vocab_file)
os.unlink(vocab_file)
tokenizer = tokenization.FullTokenizer(vocab_file)
os.unlink(vocab_file)
tokens = tokenizer.tokenize(u"UNwant\u00E9d,running")
self.assertAllEqual(tokens, ["un", "##want", "##ed", ",", "runn", "##ing"])
tokens = tokenizer.tokenize(u"UNwant\u00E9d,running")
self.assertAllEqual(tokens, ["un", "##want", "##ed", ",", "runn", "##ing"])
self.assertAllEqual(
tokenizer.convert_tokens_to_ids(tokens), [7, 4, 5, 10, 8, 9])
self.assertAllEqual(
tokenizer.convert_tokens_to_ids(tokens), [7, 4, 5, 10, 8, 9])
def test_basic_tokenizer_lower(self):
tokenizer = tokenization.BasicTokenizer(do_lower_case=True)
def test_basic_tokenizer_lower(self):
tokenizer = tokenization.BasicTokenizer(do_lower_case=True)
self.assertAllEqual(
tokenizer.tokenize(u" \tHeLLo!how \n Are yoU? "),
["hello", "!", "how", "are", "you", "?"])
self.assertAllEqual(tokenizer.tokenize(u"H\u00E9llo"), ["hello"])
self.assertAllEqual(
tokenizer.tokenize(u" \tHeLLo!how \n Are yoU? "),
["hello", "!", "how", "are", "you", "?"])
self.assertAllEqual(tokenizer.tokenize(u"H\u00E9llo"), ["hello"])
def test_basic_tokenizer_no_lower(self):
tokenizer = tokenization.BasicTokenizer(do_lower_case=False)
def test_basic_tokenizer_no_lower(self):
tokenizer = tokenization.BasicTokenizer(do_lower_case=False)
self.assertAllEqual(
tokenizer.tokenize(u" \tHeLLo!how \n Are yoU? "),
["HeLLo", "!", "how", "Are", "yoU", "?"])
self.assertAllEqual(
tokenizer.tokenize(u" \tHeLLo!how \n Are yoU? "),
["HeLLo", "!", "how", "Are", "yoU", "?"])
def test_wordpiece_tokenizer(self):
vocab_tokens = [
"[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn",
"##ing"
]
def test_wordpiece_tokenizer(self):
vocab_tokens = [
"[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn",
"##ing"
]
vocab = {}
for (i, token) in enumerate(vocab_tokens):
vocab[token] = i
tokenizer = tokenization.WordpieceTokenizer(vocab=vocab)
vocab = {}
for (i, token) in enumerate(vocab_tokens):
vocab[token] = i
tokenizer = tokenization.WordpieceTokenizer(vocab=vocab)
self.assertAllEqual(tokenizer.tokenize(""), [])
self.assertAllEqual(tokenizer.tokenize(""), [])
self.assertAllEqual(
tokenizer.tokenize("unwanted running"),
["un", "##want", "##ed", "runn", "##ing"])
self.assertAllEqual(
tokenizer.tokenize("unwanted running"),
["un", "##want", "##ed", "runn", "##ing"])
self.assertAllEqual(
tokenizer.tokenize("unwantedX running"), ["[UNK]", "runn", "##ing"])
self.assertAllEqual(
tokenizer.tokenize("unwantedX running"), ["[UNK]", "runn", "##ing"])
def test_convert_tokens_to_ids(self):
vocab_tokens = [
"[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn",
"##ing"
]
def test_convert_tokens_to_ids(self):
vocab_tokens = [
"[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn",
"##ing"
]
vocab = {}
for (i, token) in enumerate(vocab_tokens):
vocab[token] = i
vocab = {}
for (i, token) in enumerate(vocab_tokens):
vocab[token] = i
self.assertAllEqual(
tokenization.convert_tokens_to_ids(
vocab, ["un", "##want", "##ed", "runn", "##ing"]), [7, 4, 5, 8, 9])
self.assertAllEqual(
tokenization.convert_tokens_to_ids(
vocab, ["un", "##want", "##ed", "runn", "##ing"]), [7, 4, 5, 8, 9])
def test_is_whitespace(self):
self.assertTrue(tokenization._is_whitespace(u" "))
self.assertTrue(tokenization._is_whitespace(u"\t"))
self.assertTrue(tokenization._is_whitespace(u"\r"))
self.assertTrue(tokenization._is_whitespace(u"\n"))
self.assertTrue(tokenization._is_whitespace(u"\u00A0"))
def test_is_whitespace(self):
self.assertTrue(tokenization._is_whitespace(u" "))
self.assertTrue(tokenization._is_whitespace(u"\t"))
self.assertTrue(tokenization._is_whitespace(u"\r"))
self.assertTrue(tokenization._is_whitespace(u"\n"))
self.assertTrue(tokenization._is_whitespace(u"\u00A0"))
self.assertFalse(tokenization._is_whitespace(u"A"))
self.assertFalse(tokenization._is_whitespace(u"-"))
self.assertFalse(tokenization._is_whitespace(u"A"))
self.assertFalse(tokenization._is_whitespace(u"-"))
def test_is_control(self):
self.assertTrue(tokenization._is_control(u"\u0005"))
def test_is_control(self):
self.assertTrue(tokenization._is_control(u"\u0005"))
self.assertFalse(tokenization._is_control(u"A"))
self.assertFalse(tokenization._is_control(u" "))
self.assertFalse(tokenization._is_control(u"\t"))
self.assertFalse(tokenization._is_control(u"\r"))
self.assertFalse(tokenization._is_control(u"A"))
self.assertFalse(tokenization._is_control(u" "))
self.assertFalse(tokenization._is_control(u"\t"))
self.assertFalse(tokenization._is_control(u"\r"))
def test_is_punctuation(self):
self.assertTrue(tokenization._is_punctuation(u"-"))
self.assertTrue(tokenization._is_punctuation(u"$"))
self.assertTrue(tokenization._is_punctuation(u"`"))
self.assertTrue(tokenization._is_punctuation(u"."))
def test_is_punctuation(self):
self.assertTrue(tokenization._is_punctuation(u"-"))
self.assertTrue(tokenization._is_punctuation(u"$"))
self.assertTrue(tokenization._is_punctuation(u"`"))
self.assertTrue(tokenization._is_punctuation(u"."))
self.assertFalse(tokenization._is_punctuation(u"A"))
self.assertFalse(tokenization._is_punctuation(u" "))
self.assertFalse(tokenization._is_punctuation(u"A"))
self.assertFalse(tokenization._is_punctuation(u" "))
if __name__ == "__main__":
tf.test.main()
tf.test.main()
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment