remove append_LARS not used api test=develop (#16703)

paddle/fluid/API.spec

@@ -361,7 +361,6 @@ paddle.fluid.layers.inverse_time_decay (ArgSpec(args=['learning_rate', 'decay_st
 paddle.fluid.layers.polynomial_decay (ArgSpec(args=['learning_rate', 'decay_steps', 'end_learning_rate', 'power', 'cycle'], varargs=None, keywords=None, defaults=(0.0001, 1.0, False)), ('document', '882634f420f626642f0874481263da40'))
 paddle.fluid.layers.piecewise_decay (ArgSpec(args=['boundaries', 'values'], varargs=None, keywords=None, defaults=None), ('document', 'c717d9d1d78a53c809d01b8bc56f3cae'))
 paddle.fluid.layers.noam_decay (ArgSpec(args=['d_model', 'warmup_steps'], varargs=None, keywords=None, defaults=None), ('document', 'd9a95746353fd574be36dc28d8726c28'))
-paddle.fluid.layers.append_LARS (ArgSpec(args=['params_grads', 'learning_rate', 'weight_decay'], varargs=None, keywords=None, defaults=None), ('document', 'd24fa1e7d62ac8a534fc6a86002f84f8'))
 paddle.fluid.layers.cosine_decay (ArgSpec(args=['learning_rate', 'step_each_epoch', 'epochs'], varargs=None, keywords=None, defaults=None), ('document', '9588c64c26ffaef3c466e404a6af9d9b'))
 paddle.fluid.layers.linear_lr_warmup (ArgSpec(args=['learning_rate', 'warmup_steps', 'start_lr', 'end_lr'], varargs=None, keywords=None, defaults=None), ('document', '2ef3f5ca5cd71ea4217c418e5a7a0565'))
 paddle.fluid.contrib.InitState.__init__ (ArgSpec(args=['self', 'init', 'shape', 'value', 'init_boot', 'need_reorder', 'dtype'], varargs=None, keywords=None, defaults=(None, None, 0.0, None, False, 'float32')), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))

python/paddle/fluid/layers/learning_rate_scheduler.py

@@ -35,8 +35,8 @@ from ..dygraph import learning_rate_scheduler as imperate_lr
 
 __all__ = [
     'exponential_decay', 'natural_exp_decay', 'inverse_time_decay',
-    'polynomial_decay', 'piecewise_decay', 'noam_decay', 'append_LARS',
+    'polynomial_decay', 'piecewise_decay', 'noam_decay', 'cosine_decay',
-    'cosine_decay', 'linear_lr_warmup'
+    'linear_lr_warmup'
 ]
 
 
@@ -381,50 +381,6 @@ def cosine_decay(learning_rate, step_each_epoch, epochs):
             return decayed_lr
 
 
-def append_LARS(params_grads, learning_rate, weight_decay):
-    """
-    Applies LARS (LAYER-WISE ADAPTIVE RATE SCALING) to learning rate for
-    each layer.
-
-    Args:
-        learning_rate: A learning rate Variable. This
-          is the global learning rate for LARS.
-        weight_decay: A Python `float` number.
-
-    Returns:
-        The decayed learning rate
-    Examples:
-        .. code-block:: python
-
-            learning_rate *= local_gw_ratio * sqrt(sumsq(param))
-                        / (sqrt(sumsq(gradient))+ weight_decay * sqrt(sumsq(param)))
-    """
-
-    assert not imperative_base.enabled(
-    ), "append_LARS is NOT supported in dygraph mode now"
-
-    def _balanced_weight(param_norm, grad_norm):
-        if weight_decay == 1.0:
-            return grad_norm + param_norm
-        else:
-            return grad_norm + weight_decay * param_norm
-
-    for param, grad in params_grads:
-        with param.block.program.optimized_guard(
-            [param, grad]), name_scope("optimizer"):
-            param_lr = param.optimize_attr['learning_rate']
-            param_norm = ops.sqrt(nn.reduce_sum(input=ops.square(param)))
-            grad_norm = ops.sqrt(nn.reduce_sum(input=ops.square(grad)))
-            if type(param_lr) == float and param_lr == 1.0:
-                decayed_lr = learning_rate * param_norm \
-                    / _balanced_weight(param_norm, grad_norm)
-            else:
-                decayed_lr = learning_rate * param_lr * param_norm \
-                    / _balanced_weight(param_norm, grad_norm)
-            # set back param local learning rate
-            param.optimize_attr['learning_rate'] = decayed_lr
-
-
 def linear_lr_warmup(learning_rate, warmup_steps, start_lr, end_lr):
     """
     Applies linear learning rate warmup before the normal learning rate