SinaTools 0.1.40__py2.py3-none-any.whl → 1.0.1__py2.py3-none-any.whl

sinatools/arabert/araelectra/util/utils.py

@@ -1,109 +0,0 @@
-# coding=utf-8
-# Copyright 2020 The Google Research Authors.
-#
-# 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.
-
-"""A collection of general utility functions."""
-
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-import json
-import pickle
-import sys
-
-import tensorflow as tf
-
-
-def load_json(path):
-  with tf.io.gfile.GFile(path, "r") as f:
-    return json.load(f)
-
-
-def write_json(o, path):
-  if "/" in path:
-    tf.io.gfile.makedirs(path.rsplit("/", 1)[0])
-  with tf.io.gfile.GFile(path, "w") as f:
-    json.dump(o, f)
-
-
-def load_pickle(path):
-  with tf.io.gfile.GFile(path, "rb") as f:
-    return pickle.load(f)
-
-
-def write_pickle(o, path):
-  if "/" in path:
-    tf.io.gfile.makedirs(path.rsplit("/", 1)[0])
-  with tf.io.gfile.GFile(path, "wb") as f:
-    pickle.dump(o, f, -1)
-
-
-def mkdir(path):
-  if not tf.io.gfile.exists(path):
-    tf.io.gfile.makedirs(path)
-
-
-def rmrf(path):
-  if tf.io.gfile.exists(path):
-    tf.io.gfile.rmtree(path)
-
-
-def rmkdir(path):
-  rmrf(path)
-  mkdir(path)
-
-
-def log(*args):
-  msg = " ".join(map(str, args))
-  sys.stdout.write(msg + "\n")
-  sys.stdout.flush()
-
-
-def log_config(config):
-  for key, value in sorted(config.__dict__.items()):
-    log(key, value)
-  log()
-
-
-def heading(*args):
-  log(80 * "=")
-  log(*args)
-  log(80 * "=")
-
-
-def nest_dict(d, prefixes, delim="_"):
-  """Go from {prefix_key: value} to {prefix: {key: value}}."""
-  nested = {}
-  for k, v in d.items():
-    for prefix in prefixes:
-      if k.startswith(prefix + delim):
-        if prefix not in nested:
-          nested[prefix] = {}
-        nested[prefix][k.split(delim, 1)[1]] = v
-      else:
-        nested[k] = v
-  return nested
-
-
-def flatten_dict(d, delim="_"):
-  """Go from {prefix: {key: value}} to {prefix_key: value}."""
-  flattened = {}
-  for k, v in d.items():
-    if isinstance(v, dict):
-      for k2, v2 in v.items():
-        flattened[k + delim + k2] = v2
-    else:
-      flattened[k] = v
-  return flattened

sinatools/arabert/aragpt2/__init__.py

@@ -1,2 +0,0 @@
-# coding=utf-8
-

sinatools/arabert/aragpt2/create_pretraining_data.py

@@ -1,95 +0,0 @@
-import collections
-from transformers import GPT2TokenizerFast
-import tensorflow as tf
-
-import sys
-sys.path.append("..")
-from arabert.preprocess import preprocess
-
-flags = tf.flags
-
-FLAGS = flags.FLAGS
-
-flags.DEFINE_string(
-    "input_file", None, "Input raw text file (or comma-separated list of files)."
-)
-
-flags.DEFINE_string(
-    "output_file", None, "Output TF example file (or comma-separated list of files)."
-)
-
-flags.DEFINE_string(
-    "tokenizer_dir", None, "The directory of a pretrained GPT2TokenizerFast"
-)
-
-flags.DEFINE_integer(
-    "max_len", 1024, "The vocabulary file that the BERT model was trained on."
-)
-
-flags.DEFINE_integer(
-    "num_examples_print", 0, "Number of examples to print"
-)
-
-
-def create_int_feature(values):
-    feature = tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))
-    return feature
-
-
-def main(_):
-    tf.logging.set_verbosity(tf.logging.INFO)
-    logger = tf.get_logger()
-    logger.propagate = False
-
-    input_files = []
-    for input_pattern in FLAGS.input_file.split(","):
-        input_files.extend(tf.gfile.Glob(input_pattern))
-
-    tf.logging.info("*** Reading from input files ***")
-    for input_file in input_files:
-        tf.logging.info("  %s", input_file)
-
-    gpt2_tok = GPT2TokenizerFast.from_pretrained(FLAGS.tokenizer_dir)
-
-    writer = tf.python_io.TFRecordWriter(FLAGS.output_file + ".tfrecord")
-
-    eos_id = gpt2_tok.eos_token_id
-    all_examples = []
-    for input_file in input_files:
-        queue = []
-        example = []
-        with tf.gfile.GFile(input_file, "r") as reader:
-            for line in reader.readlines():
-                if line == "\n":
-                    queue.append(eos_id)
-                else:
-                    line = line.replace("\n", " ")
-                    line = preprocess(line,model='gpt2-base-arabic')
-                    line = line.strip()
-                    enc_line = gpt2_tok.encode(line)
-                    queue.extend(enc_line)
-                if len(queue) > FLAGS.max_len +1:
-                    example = [queue.pop(0) for _ in range(FLAGS.max_len +1)]
-                    assert len(example) == FLAGS.max_len +1
-                    all_examples.append(example)
-
-
-    for i, ex in enumerate(all_examples):
-        features = collections.OrderedDict()
-        features["input_ids"] = create_int_feature(ex)
-        tf_example = tf.train.Example(features=tf.train.Features(feature=features))
-        writer.write(tf_example.SerializeToString())
-        if i < FLAGS.num_examples_print:
-            tf.logging.info("*** Example ***")
-            tf.logging.info("Length: %d" % len(ex))
-            tf.logging.info("Tokens: %s" % gpt2_tok.decode(ex))
-            tf.logging.info("ids: %s" % " ".join([str(x) for x in ex]))
-
-    tf.logging.info("Wrote %d total instances", len(all_examples))
-
-
-if __name__ == "__main__":
-    flags.mark_flag_as_required("input_file")
-    flags.mark_flag_as_required("output_file")
-    flags.mark_flag_as_required("tokenizer_dir")
-    tf.app.run()

sinatools/arabert/aragpt2/gpt2/__init__.py

@@ -1,2 +0,0 @@
-# coding=utf-8
-

sinatools/arabert/aragpt2/gpt2/lamb_optimizer.py

@@ -1,158 +0,0 @@
-# coding=utf-8
-# Copyright 2019 The Google Research Authors.
-#
-# 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.
-
-# Lint as: python2, python3
-"""Functions and classes related to optimization (weight updates)."""
-
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-import re
-import six
-import tensorflow as tf
-
-# pylint: disable=g-direct-tensorflow-import
-from tensorflow.python.ops import array_ops
-from tensorflow.python.ops import linalg_ops
-from tensorflow.python.ops import math_ops
-
-# pylint: enable=g-direct-tensorflow-import
-
-
-class LAMBOptimizer(tf.train.Optimizer):
-    """LAMB (Layer-wise Adaptive Moments optimizer for Batch training)."""
-
-    # A new optimizer that includes correct L2 weight decay, adaptive
-    # element-wise updating, and layer-wise justification. The LAMB optimizer
-    # was proposed by Yang You, Jing Li, Jonathan Hseu, Xiaodan Song,
-    # James Demmel, and Cho-Jui Hsieh in a paper titled as Reducing BERT
-    # Pre-Training Time from 3 Days to 76 Minutes (arxiv.org/abs/1904.00962)
-
-    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,
-        exclude_from_layer_adaptation=None,
-        name="LAMBOptimizer",
-    ):
-        """Constructs a LAMBOptimizer."""
-        super(LAMBOptimizer, 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
-        # exclude_from_layer_adaptation is set to exclude_from_weight_decay if the
-        # arg is None.
-        # TODO(jingli): validate if exclude_from_layer_adaptation is necessary.
-        if exclude_from_layer_adaptation:
-            self.exclude_from_layer_adaptation = exclude_from_layer_adaptation
-        else:
-            self.exclude_from_layer_adaptation = 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=six.ensure_str(param_name) + "/adam_m",
-                shape=param.shape.as_list(),
-                dtype=tf.float32,
-                trainable=False,
-                initializer=tf.zeros_initializer(),
-            )
-            v = tf.get_variable(
-                name=six.ensure_str(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
-
-            ratio = 1.0
-            if self._do_layer_adaptation(param_name):
-                w_norm = linalg_ops.norm(param, ord=2)
-                g_norm = linalg_ops.norm(update, ord=2)
-                ratio = array_ops.where(
-                    math_ops.greater(w_norm, 0),
-                    array_ops.where(
-                        math_ops.greater(g_norm, 0), (w_norm / g_norm), 1.0
-                    ),
-                    1.0,
-                )
-
-            update_with_lr = ratio * 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 _do_layer_adaptation(self, param_name):
-        """Whether to do layer-wise learning rate adaptation for `param_name`."""
-        if self.exclude_from_layer_adaptation:
-            for r in self.exclude_from_layer_adaptation:
-                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+$", six.ensure_str(param_name))
-        if m is not None:
-            param_name = m.group(1)
-        return param_name

sinatools/arabert/aragpt2/gpt2/optimization.py

@@ -1,225 +0,0 @@
-# coding=utf-8
-# Copyright 2018 The Google AI Language Team Authors.
-#
-# 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.
-"""Functions and classes related to optimization (weight updates)."""
-
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-import re
-import tensorflow as tf
-import lamb_optimizer
-from gpt_2_simple.src import memory_saving_gradients
-
-
-def create_optimizer(
-    loss,
-    init_lr,
-    num_train_steps,
-    num_warmup_steps,
-    use_tpu,
-    optimizer="lamb",
-    poly_power=1.0,
-    start_warmup_step=0,
-    use_memory_saving_gradients=False
-):
-    """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)
-
-    # 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=poly_power,
-        cycle=False,
-    )
-
-    # Implements linear warmup. I.e., if global_step - start_warmup_step <
-    # num_warmup_steps, the learning rate will be
-    # `(global_step - start_warmup_step)/num_warmup_steps * init_lr`.
-    if num_warmup_steps:
-        tf.logging.info(
-            "++++++ warmup starts at step "
-            + str(start_warmup_step)
-            + ", for "
-            + str(num_warmup_steps)
-            + " steps ++++++"
-        )
-        global_steps_int = tf.cast(global_step, tf.int32)
-        start_warm_int = tf.constant(start_warmup_step, dtype=tf.int32)
-        global_steps_int = global_steps_int - start_warm_int
-        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)
-
-        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
-
-    # It is OK that you use this optimizer for finetuning, since this
-    # is how the model was trained (note that the Adam m/v variables are NOT
-    # loaded from init_checkpoint.)
-    # It is OK to use AdamW in the finetuning even the model is trained by LAMB.
-    # As report in the Bert pulic github, the learning rate for SQuAD 1.1 finetune
-    # is 3e-5, 4e-5 or 5e-5. For LAMB, the users can use 3e-4, 4e-4,or 5e-4 for a
-    # batch size of 64 in the finetune.
-    if optimizer == "adamw":
-        tf.logging.info("using adamw")
-        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"],
-        )
-    elif optimizer == "lamb":
-        tf.logging.info("using lamb")
-        optimizer = lamb_optimizer.LAMBOptimizer(
-            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"],
-        )
-    else:
-        raise ValueError("Not supported optimizer: ", optimizer)
-
-    if use_tpu:
-        optimizer = tf.contrib.tpu.CrossShardOptimizer(optimizer)
-
-    tvars = tf.trainable_variables()
-
-
-    if use_memory_saving_gradients:
-        grads = memory_saving_gradients.gradients(loss, tvars)
-    else:
-        grads = tf.gradients(ys=loss, xs=tvars)
-
-
-    # This is how bert 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)
-
-    # Normally the global step update is done inside of `apply_gradients`.
-    # However, neither `AdamWeightDecayOptimizer` nor `LAMBOptimizer` do this.
-    # But if you use a different optimizer, you should probably take this line
-    # out.
-    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