optimization.py

# coding=utf-8
# Copyright (c) 2019 NVIDIA CORPORATION. All rights reserved.
# 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
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import linalg_ops
from tensorflow.python.ops import math_ops
from horovod.tensorflow.compression import Compression

def create_optimizer(loss, init_lr, num_train_steps, num_warmup_steps, hvd=None, manual_fp16=False, use_fp16=False, num_accumulation_steps=1,
                     optimizer_type="adam", allreduce_post_accumulation=False, init_loss_scale=2**32):
  """Creates an optimizer training op."""
  global_step = tf.compat.v1.train.get_or_create_global_step()
  
  # avoid step change in learning rate at end of warmup phase
  if optimizer_type == "adam":
      power = 1.0
      decayed_learning_rate_at_crossover_point = init_lr * (
                  (1.0 - float(num_warmup_steps) / float(num_train_steps)) ** power)
  else:
      power = 0.5
      decayed_learning_rate_at_crossover_point = init_lr

  adjusted_init_lr = init_lr * (init_lr / decayed_learning_rate_at_crossover_point)
  print('decayed_learning_rate_at_crossover_point = %e, adjusted_init_lr = %e' % (decayed_learning_rate_at_crossover_point, adjusted_init_lr))

  learning_rate = tf.constant(value=adjusted_init_lr, shape=[], dtype=tf.float32)

  # Implements linear decay of the learning rate.
  learning_rate = tf.compat.v1.train.polynomial_decay(
      learning_rate,
      global_step,
      num_train_steps,
      end_learning_rate=0.0,
      power=power,
      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)

    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)

  if optimizer_type == "lamb":
      print("Initializing LAMB Optimizer")
      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:
      print("Initializing ADAM Weight Decay Optimizer")
      # 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 hvd is not None and (num_accumulation_steps == 1 or (not allreduce_post_accumulation)):
    optimizer = hvd.DistributedOptimizer(optimizer, sparse_as_dense=True, compression=Compression.fp16 if use_fp16 or manual_fp16 else Compression.none)
  if use_fp16:
    loss_scaler = tf.train.experimental.DynamicLossScale(initial_loss_scale=init_loss_scale, increment_period=1000, multiplier=2.0)
    optimizer = tf.train.experimental.enable_mixed_precision_graph_rewrite(optimizer, loss_scaler)
    loss_scale_value = tf.identity(loss_scaler(), name="loss_scale")
  if manual_fp16:
    loss_scale_manager = tf.contrib.mixed_precision.ExponentialUpdateLossScaleManager(init_loss_scale=init_loss_scale,
                                                                                    incr_every_n_steps=1000,
                                                                                    decr_every_n_nan_or_inf=2,
                                                                                    decr_ratio=0.5)
    optimizer = tf.contrib.mixed_precision.LossScaleOptimizer(optimizer, loss_scale_manager)

  tvars = tf.trainable_variables()
  grads_and_vars = optimizer.compute_gradients(loss * 1.0 / num_accumulation_steps, tvars)

  if num_accumulation_steps > 1:
      local_step = tf.get_variable(name="local_step", shape=[], dtype=tf.int32, trainable=False,
                                   initializer=tf.zeros_initializer)
      batch_finite = tf.get_variable(name="batch_finite", shape=[], dtype=tf.bool, trainable=False,
                                     initializer=tf.ones_initializer)
      accum_vars = [tf.get_variable(
          name=tvar.name.split(":")[0] + "/accum",
          shape=tvar.shape.as_list(),
          dtype=tf.float32,
          trainable=False,
          initializer=tf.zeros_initializer()) for tvar in tf.trainable_variables()]

      reset_step = tf.cast(tf.math.equal(local_step % num_accumulation_steps, 0), dtype=tf.bool)
      local_step = tf.cond(reset_step, lambda:local_step.assign(tf.ones_like(local_step)), lambda:local_step.assign_add(1))

      grads_and_vars_and_accums = [(gv[0],gv[1],accum_vars[i]) for i, gv in enumerate(grads_and_vars) if gv[0] is not None]
      grads, tvars, accum_vars = list(zip(*grads_and_vars_and_accums))

      all_are_finite = tf.reduce_all([tf.reduce_all(tf.is_finite(g)) for g in grads]) if manual_fp16 or use_fp16 else tf.constant(True, dtype=tf.bool)
      batch_finite = tf.cond(reset_step,
        lambda: batch_finite.assign(tf.math.logical_and(tf.constant(True, dtype=tf.bool), all_are_finite)),
        lambda:batch_finite.assign(tf.math.logical_and(batch_finite, all_are_finite)))

      # This is how the model was pre-trained.
      # ensure global norm is a finite number
      # to prevent clip_by_global_norm from having a hizzy fit.
      (clipped_grads, _) = tf.clip_by_global_norm(
            grads, clip_norm=1.0,
            use_norm=tf.cond(
                all_are_finite,
                lambda: tf.global_norm(grads),
                lambda: tf.constant(1.0)))

      accum_vars = tf.cond(reset_step,
              lambda: [accum_vars[i].assign(grad) for i, grad in enumerate(clipped_grads)],
              lambda: [accum_vars[i].assign_add(grad) for i, grad in enumerate(clipped_grads)])

      def update(accum_vars):
          if allreduce_post_accumulation and hvd is not None:
              accum_vars = [hvd.allreduce(tf.convert_to_tensor(accum_var), compression=Compression.fp16 if use_fp16 or manual_fp16 else Compression.none) if isinstance(accum_var, tf.IndexedSlices)
                            else hvd.allreduce(accum_var, compression=Compression.fp16 if use_fp16 or manual_fp16 else Compression.none) for accum_var in accum_vars]
          return optimizer.apply_gradients(list(zip(accum_vars, tvars)), global_step=global_step)

      update_step = tf.identity(tf.cast(tf.math.equal(local_step % num_accumulation_steps, 0), dtype=tf.bool), name="update_step")
      update_op = tf.cond(update_step,
                          lambda: update(accum_vars), lambda: tf.no_op())

      new_global_step = tf.cond(tf.math.logical_and(update_step, 
                                                    tf.cast(hvd.allreduce(tf.cast(batch_finite, tf.int32)), tf.bool) if hvd is not None else batch_finite),
                                lambda: global_step+1,
                                lambda: global_step)
      new_global_step = tf.identity(new_global_step, name='step_update')
      train_op = tf.group(update_op, [global_step.assign(new_global_step)])
  else:
      grads_and_vars = [(g, v) for g, v in grads_and_vars if g is not None]
      grads, tvars = list(zip(*grads_and_vars))
      all_are_finite = tf.reduce_all(
          [tf.reduce_all(tf.is_finite(g)) for g in grads]) if use_fp16 or manual_fp16 else tf.constant(True, dtype=tf.bool)

      # This is how the model was pre-trained.
      # ensure global norm is a finite number
      # to prevent clip_by_global_norm from having a hizzy fit.
      (clipped_grads, _) = tf.clip_by_global_norm(
          grads, clip_norm=1.0,
          use_norm=tf.cond(
              all_are_finite,
              lambda: tf.global_norm(grads),
              lambda: tf.constant(1.0)))

      train_op = optimizer.apply_gradients(
          list(zip(clipped_grads, tvars)), global_step=global_step)

      new_global_step = tf.cond(all_are_finite, lambda: global_step + 1, lambda: global_step)
      new_global_step = tf.identity(new_global_step, name='step_update')
      train_op = tf.group(train_op, [global_step.assign(new_global_step)])
  return train_op


class AdamWeightDecayOptimizer(tf.compat.v1.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 = tf.identity(learning_rate, name='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,
      manual_fp16=False):
    """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)
      has_shadow = manual_fp16 and param.dtype.base_dtype != tf.float32
      if has_shadow:
        # create shadow fp32 weights for fp16 variable
        param_fp32 = tf.get_variable(
            name=param_name + "/shadow",
            dtype=tf.float32,
            trainable=False,
            initializer=tf.cast(param.initialized_value(),tf.float32))
      else:
        param_fp32 = param

      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 to 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_fp32

      update_with_lr = self.learning_rate * update

      next_param = param_fp32 - update_with_lr

      if has_shadow:
        # cast shadow fp32 weights to fp16 and assign to trainable variable
        param.assign(tf.cast(next_param, param.dtype.base_dtype))
      assignments.extend(
          [param_fp32.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


class LAMBOptimizer(tf.compat.v1.train.Optimizer):
  """A LAMB 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="LAMBOptimizer"):
    """Constructs a LAMBOptimizer."""
    super(LAMBOptimizer, self).__init__(False, name)

    self.learning_rate = tf.identity(learning_rate, name='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, name=None,
      manual_fp16=False):
    """See base class."""
    assignments = []
    steps = tf.cast(global_step, tf.float32)
    for (grad, param) in grads_and_vars:
      if grad is None or param is None:
        continue

      param_name = self._get_variable_name(param.name)
      has_shadow = manual_fp16 and param.dtype.base_dtype != tf.float32
      if has_shadow:
        # create shadow fp32 weights for fp16 variable
        param_fp32 = tf.get_variable(
            name=param_name + "/shadow",
            dtype=tf.float32,
            trainable=False,
            initializer=tf.cast(param.initialized_value(),tf.float32))
      else:
        param_fp32 = param

      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())

      # LAMB 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)))

      beta1_correction = (1 - self.beta_1 ** steps)
      beta2_correction = (1 - self.beta_2 ** steps)

      next_m_unbiased = next_m / beta1_correction
      next_v_unbiased = next_v / beta2_correction

      update = next_m_unbiased / (tf.sqrt(next_v_unbiased) + 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 to 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_fp32

      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_fp32 - update_with_lr

      if has_shadow:
        # cast shadow fp32 weights to fp16 and assign to trainable variable
        param.assign(tf.cast(next_param, param.dtype.base_dtype))
      assignments.extend(
          [param_fp32.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