[Enhancement]Revise docstring and UT for layer decay LR (#1540)

* fix docstring

* fix ut for optimizer cosntructor

mmseg/core/optimizers/layer_decay_optimizer_constructor.py

@@ -17,7 +17,7 @@ def get_layer_id_for_convnext(var_name, max_layer_id):
         max_layer_id (int): Maximum number of backbone layers.
 
     Returns:
-        int: The id number corresponding to different　learning rate in
+        int: The id number corresponding to different learning rate in
         ``LearningRateDecayOptimizerConstructor``.
     """
 
@@ -60,7 +60,7 @@ def get_stage_id_for_convnext(var_name, max_stage_id):
         max_stage_id (int): Maximum number of backbone layers.
 
     Returns:
-        int: The id number corresponding to different　learning rate in
+        int: The id number corresponding to different learning rate in
         ``LearningRateDecayOptimizerConstructor``.
     """
 
@@ -103,8 +103,8 @@ def get_layer_id_for_vit(var_name, max_layer_id):
 class LearningRateDecayOptimizerConstructor(DefaultOptimizerConstructor):
     """Different learning rates are set for different layers of backbone.
 
-    Note: Currently, this optimizer constructor is built for ConvNeXt
-    and BEiT.
+    Note: Currently, this optimizer constructor is built for ConvNeXt,
+    BEiT and MAE.
     """
 
     def add_params(self, params, module, **kwargs):

tests/test_core/test_layer_decay_optimizer_constructor.py

@@ -157,6 +157,19 @@ class ToyBEiT(nn.Module):
             self.layers.append(layer)
 
 
+class ToyMAE(nn.Module):
+
+    def __init__(self):
+        super().__init__()
+        # add some variables to meet unit test coverate rate
+        self.cls_token = nn.Parameter(torch.ones(1))
+        self.patch_embed = nn.Parameter(torch.ones(1))
+        self.layers = nn.ModuleList()
+        for _ in range(3):
+            layer = nn.Conv2d(3, 3, 1)
+            self.layers.append(layer)
+
+
 class ToySegmentor(nn.Module):
 
     def __init__(self, backbone):
@@ -236,6 +249,17 @@ def test_learning_rate_decay_optimizer_constructor():
             optimizer_cfg, stagewise_paramwise_cfg)
         optimizer = optim_constructor(model)
 
+    # Test lr wd for MAE
+    backbone = ToyMAE()
+    model = PseudoDataParallel(ToySegmentor(backbone))
+
+    layerwise_paramwise_cfg = dict(
+        decay_rate=decay_rate, decay_type='layer_wise', num_layers=3)
+    optim_constructor = LearningRateDecayOptimizerConstructor(
+        optimizer_cfg, layerwise_paramwise_cfg)
+    optimizer = optim_constructor(model)
+    check_optimizer_lr_wd(optimizer, expected_layer_wise_wd_lr_beit)
+
 
 def test_beit_layer_decay_optimizer_constructor():