更换优化器出现OOM问题

请问一下，在调用Bert optimization中自定义的优化器时出现OOM，更换成tf.train.AdamOptimizer就OK了，是什么原因？
代码如下：

with tf.variable_scope("optimizer"):
    num_warmup_steps = int(num_train_steps * config['warmup_proportion'])
    # 导入bert自定义optimization
    self.train_op = optimization.create_optimizer(self.loss, self.lr, self.num_train_steps,
                                             num_warmup_steps, False)

出现OOM，更换成以下优化器就OK了

with tf.variable_scope("optimizer"):
    self.opt = tf.train.AdamOptimizer(learning_rate)
    # 梯度截断
    grads = tf.gradients(self.loss, train_vars)
    (grads, _) = tf.clip_by_global_norm(grads, clip_norm=1.0)
    self.train_op = self.opt.apply_gradients(
        zip(grads, train_vars), global_step=self.global_step
    )