diff --git a/pyproject.toml b/pyproject.toml
index f244f6c1d45e4..da0f272d4602b 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -54,6 +54,10 @@ select = [
 
     # NumPy-specific rules
     "NPY001",
+
+    # Bugbear
+    "B009",
+    "B010",
 ]
 unfixable = [
     "NPY001"
diff --git a/python/paddle/dataset/common.py b/python/paddle/dataset/common.py
index eab4d37676190..fb8c4ba969164 100644
--- a/python/paddle/dataset/common.py
+++ b/python/paddle/dataset/common.py
@@ -138,10 +138,7 @@ def fetch_all():
         if "fetch" in dir(
             importlib.import_module("paddle.dataset.%s" % module_name)
         ):
-            getattr(
-                importlib.import_module("paddle.dataset.%s" % module_name),
-                "fetch",
-            )()
+            importlib.import_module('paddle.dataset.%s' % module_name).fetch()
 
 
 def split(reader, line_count, suffix="%05d.pickle", dumper=pickle.dump):
diff --git a/python/paddle/distributed/fleet/fleet.py b/python/paddle/distributed/fleet/fleet.py
index 96a964b6999cd..55650a5b5f6af 100755
--- a/python/paddle/distributed/fleet/fleet.py
+++ b/python/paddle/distributed/fleet/fleet.py
@@ -1282,7 +1282,7 @@ def _minimize_impl(
             self.origin_main_program = loss.block.program
             # add distributed attr
             if not hasattr(self.origin_main_program, "distributed_info_"):
-                setattr(self.origin_main_program, "distributed_info_", dict())
+                self.origin_main_program.distributed_info_ = dict()
                 self.origin_main_program.distributed_info_[
                     "dp_degree"
                 ] = self._user_defined_strategy.sharding_configs["dp_degree"]
diff --git a/python/paddle/distributed/fleet/layers/mpu/mp_layers.py b/python/paddle/distributed/fleet/layers/mpu/mp_layers.py
index 794acde2ec88a..c8914256021b8 100644
--- a/python/paddle/distributed/fleet/layers/mpu/mp_layers.py
+++ b/python/paddle/distributed/fleet/layers/mpu/mp_layers.py
@@ -143,7 +143,7 @@ def __init__(
 
         self.weight.is_distributed = True if self.is_mp else False
         if self.weight.is_distributed:
-            setattr(self.weight, "split_axis", 0)
+            self.weight.split_axis = 0
 
     def forward(self, x):
         if self.is_mp:
@@ -277,7 +277,7 @@ def __init__(
         self.weight.is_distributed = True if self.is_mp else False
 
         if self.weight.is_distributed:
-            setattr(self.weight, "split_axis", 1)
+            self.weight.split_axis = 1
 
         if has_bias:
             # initialize bias to zero like Megatron
@@ -289,7 +289,7 @@ def __init__(
             )
             self.bias.is_distributed = True if self.is_mp else False
             if self.bias.is_distributed:
-                setattr(self.bias, "split_axis", 0)
+                self.bias.split_axis = 0
         else:
             self.bias = None
 
@@ -443,7 +443,7 @@ def __init__(
 
         self.weight.is_distributed = True if self.is_mp else False
         if self.weight.is_distributed:
-            setattr(self.weight, "split_axis", 0)
+            self.weight.split_axis = 0
 
         if has_bias:
             self.bias = self.create_parameter(
diff --git a/python/paddle/distributed/fleet/meta_parallel/parallel_layers/pp_layers.py b/python/paddle/distributed/fleet/meta_parallel/parallel_layers/pp_layers.py
index 3b164d2afa1cb..3c1d615ed63f1 100755
--- a/python/paddle/distributed/fleet/meta_parallel/parallel_layers/pp_layers.py
+++ b/python/paddle/distributed/fleet/meta_parallel/parallel_layers/pp_layers.py
@@ -493,7 +493,7 @@ def _synchronize_shared_weights(self):
 
             for param in comm['layer'].parameters():
                 if self.global_rank != min(comm['ranks']):
-                    setattr(param, 'is_firstly_shared', False)
+                    param.is_firstly_shared = False
 
     def allreduce_shared_weight_gradients(self):
         for key, comm in self.shared_comm.items():
@@ -641,7 +641,7 @@ def _build_layer_impl(self, start, end):
                     for param in self.shared_layers[
                         layer.layer_name
                     ].parameters():
-                        setattr(param, "is_firstly_shared", True)
+                        param.is_firstly_shared = True
 
                 if layer.forward_func is None:
                     run_function.append(self.shared_layers[layer.layer_name])
diff --git a/python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_stage3.py b/python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_stage3.py
index 5fb2e9a58d501..9e440f3fe671b 100644
--- a/python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_stage3.py
+++ b/python/paddle/distributed/fleet/meta_parallel/sharding/group_sharded_stage3.py
@@ -1047,18 +1047,18 @@ def _create_params_grad(trainable_params, param2buffer_size, task_flow):
 
 def _PartitionParam(param):
     if not hasattr(param, "fw_storage"):
-        setattr(param, "fw_storage", None)
-        setattr(param, "bw_storage", None)
-        setattr(param, "master_weight", None)
-        setattr(param, "status", "all")
-        setattr(param, "use_count", 0)
+        param.fw_storage = None
+        param.bw_storage = None
+        param.master_weight = None
+        param.status = "all"
+        param.use_count = 0
     return param
 
 
 def _UnsliceParam(param):
     if not hasattr(param, "unslice"):
-        setattr(param, "unslice", True)
-        setattr(param, "master_weight", None)
+        param.unslice = True
+        param.master_weight = None
     return param
 
 
@@ -1078,11 +1078,11 @@ def _VarBaseWrapper(param):
 
 def _OptimizerWrapper(optimizer, offload, group, update_params_slice):
     if not hasattr(optimizer, "_optim"):
-        setattr(optimizer, "_optim", optimizer)
-        setattr(optimizer, "offload", offload)
-        setattr(optimizer, "_group", group)
-        setattr(optimizer, "update_scaler", None)
-        setattr(optimizer, "update_slice", update_params_slice)
+        optimizer._optim = optimizer
+        optimizer.offload = offload
+        optimizer._group = group
+        optimizer.update_scaler = None
+        optimizer.update_slice = update_params_slice
     return optimizer
 
 
diff --git a/python/paddle/fluid/tests/custom_op/test_custom_raw_op_kernel_op.py b/python/paddle/fluid/tests/custom_op/test_custom_raw_op_kernel_op.py
index 3fcf94e2f1fda..9762d29c48c83 100644
--- a/python/paddle/fluid/tests/custom_op/test_custom_raw_op_kernel_op.py
+++ b/python/paddle/fluid/tests/custom_op/test_custom_raw_op_kernel_op.py
@@ -67,7 +67,7 @@ def tearDownClass(cls):
 
     def custom_raw_relu(self, x):
         module = importlib.import_module(MODULE_NAME)
-        custom_raw_relu_op = getattr(module, "custom_raw_relu")
+        custom_raw_relu_op = module.custom_raw_relu
         self.assertIsNotNone(custom_raw_relu_op)
         return custom_raw_relu_op(x)
 
diff --git a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_convert_operators.py b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_convert_operators.py
index 812abb18ff356..895203946018f 100644
--- a/python/paddle/fluid/tests/unittests/dygraph_to_static/test_convert_operators.py
+++ b/python/paddle/fluid/tests/unittests/dygraph_to_static/test_convert_operators.py
@@ -31,7 +31,7 @@ def forward(self):
 
 
 net = ForwardNotExist()
-setattr(net, "forward", "A string so that convert forward will fail")
+net.forward = "A string so that convert forward will fail"
 
 
 class TestConvertCall(unittest.TestCase):
diff --git a/python/paddle/fluid/tests/unittests/eager_op_test.py b/python/paddle/fluid/tests/unittests/eager_op_test.py
index 9615a9c81531a..3aeb0bc87ce4e 100644
--- a/python/paddle/fluid/tests/unittests/eager_op_test.py
+++ b/python/paddle/fluid/tests/unittests/eager_op_test.py
@@ -449,7 +449,7 @@ def is_bfloat16_op(self):
             )
             or (
                 hasattr(self, 'mkldnn_data_type')
-                and getattr(self, 'mkldnn_data_type') == "bfloat16"
+                and self.mkldnn_data_type == "bfloat16"
             )
             or (
                 hasattr(self, 'attrs')
@@ -469,7 +469,7 @@ def is_float16_op(self):
             )
             or (
                 hasattr(self, 'mkldnn_data_type')
-                and getattr(self, 'mkldnn_data_type') == "float16"
+                and self.mkldnn_data_type == "float16"
             )
             or (
                 hasattr(self, 'attrs')
@@ -1713,7 +1713,7 @@ def _is_skip_name(self, name):
             prim_checker = PrimForwardChecker(self, place)
             prim_checker.check()
             # Support operators which are not in the NO_FP64_CHECK_GRAD_OP_LIST list can be test prim with fp32
-            setattr(self.__class__, 'check_prim', True)
+            self.__class__.check_prim = True
             self.__class__.op_type = self.op_type
         # set some flags by the combination of arguments.
         self.infer_dtype_from_inputs_outputs(self.inputs, self.outputs)
@@ -1728,8 +1728,9 @@ def _is_skip_name(self, name):
             if self.is_mkldnn_op():
                 check_dygraph = False
 
-                if hasattr(self, 'force_fp32_output') and getattr(
-                    self, 'force_fp32_output'
+                if (
+                    hasattr(self, 'force_fp32_output')
+                    and self.force_fp32_output
                 ):
                     atol = 1e-2 if atol < 1e-2 else atol
                 else:
@@ -2078,7 +2079,7 @@ def check_grad_with_place(
             )
             prim_grad_checker.check()
             # Support operators which are not in the NO_FP64_CHECK_GRAD_OP_LIST list can be test prim with fp32
-            setattr(self.__class__, 'check_prim', True)
+            self.__class__.check_prim = True
             self._check_grad_helper()
             if only_check_prim:
                 return
diff --git a/python/paddle/fluid/tests/unittests/op_test.py b/python/paddle/fluid/tests/unittests/op_test.py
index 74e35abe5d29c..cdb521fe258f6 100644
--- a/python/paddle/fluid/tests/unittests/op_test.py
+++ b/python/paddle/fluid/tests/unittests/op_test.py
@@ -451,7 +451,7 @@ def is_bfloat16_op(self):
             )
             or (
                 hasattr(self, 'mkldnn_data_type')
-                and getattr(self, 'mkldnn_data_type') == "bfloat16"
+                and self.mkldnn_data_type == "bfloat16"
             )
             or (
                 hasattr(self, 'attrs')
@@ -471,7 +471,7 @@ def is_float16_op(self):
             )
             or (
                 hasattr(self, 'mkldnn_data_type')
-                and getattr(self, 'mkldnn_data_type') == "float16"
+                and self.mkldnn_data_type == "float16"
             )
             or (
                 hasattr(self, 'attrs')
@@ -1502,7 +1502,7 @@ def check_output_with_place(
             prim_checker = PrimForwardChecker(self, place)
             prim_checker.check()
             # Support operators which not in the NO_FP64_CHECK_GRAD_OP_LIST list can be test prim with fp32
-            setattr(self.__class__, 'check_prim', True)
+            self.__class__.check_prim = True
             self.__class__.op_type = self.op_type
         # disable legacy dygraph check when check_eager is True
         if check_eager:
@@ -1907,8 +1907,9 @@ def _is_skip_name(self, name):
             if self.is_mkldnn_op():
                 check_dygraph = False
                 check_eager = False
-                if hasattr(self, 'force_fp32_output') and getattr(
-                    self, 'force_fp32_output'
+                if (
+                    hasattr(self, 'force_fp32_output')
+                    and self.force_fp32_output
                 ):
                     atol = 1e-2 if atol < 1e-2 else atol
                 else:
@@ -2288,7 +2289,7 @@ def check_grad_with_place(
             )
             prim_grad_checker.check()
             # Support operators which not in the NO_FP64_CHECK_GRAD_OP_LIST list can be test prim with fp32
-            setattr(self.__class__, 'check_prim', True)
+            self.__class__.check_prim = True
             self._check_grad_helper()
             if only_check_prim:
                 return
diff --git a/python/paddle/hapi/model_summary.py b/python/paddle/hapi/model_summary.py
index d6234eacd15af..c36b9552be4aa 100644
--- a/python/paddle/hapi/model_summary.py
+++ b/python/paddle/hapi/model_summary.py
@@ -312,8 +312,8 @@ def hook(layer, input, output):
                 params += np.prod(v.shape)
 
                 try:
-                    if (getattr(getattr(layer, k), 'trainable')) and (
-                        not getattr(getattr(layer, k), 'stop_gradient')
+                    if (getattr(layer, k).trainable) and (
+                        not getattr(layer, k).stop_gradient
                     ):
                         summary[m_key]["trainable_params"] += np.prod(v.shape)
                         summary[m_key]["trainable"] = True
diff --git a/python/paddle/incubate/distributed/utils/io/save_for_auto.py b/python/paddle/incubate/distributed/utils/io/save_for_auto.py
index 3008201d2fd11..701c22e50fc76 100644
--- a/python/paddle/incubate/distributed/utils/io/save_for_auto.py
+++ b/python/paddle/incubate/distributed/utils/io/save_for_auto.py
@@ -219,7 +219,7 @@ def _get_dims_mapping(dist_parameter, mp_group):
 
     dist_shape = np.array(dist_parameter.shape)
     if hasattr(dist_parameter, "split_axis"):
-        aixs = getattr(dist_parameter, "split_axis")
+        aixs = dist_parameter.split_axis
         mapping = [-1 for _ in dist_shape]
         mapping[aixs] = 1
         logger.debug(
@@ -351,7 +351,7 @@ def _get_wrapped_dist_state_dict(dist_state_dict):
             logger.debug(f"not first used : {v.name}")
             continue
         wrapped_state_dict[name_mapping[v.name]] = v
-        setattr(v, "dims_mapping", _get_dims_mapping(v, mp_group))
+        v.dims_mapping = _get_dims_mapping(v, mp_group)
         logger.debug(
             f"saving param: {v.name} -> {name_mapping[v.name]} shape: {v.shape}"
         )
diff --git a/python/paddle/jit/dy2static/convert_call_func.py b/python/paddle/jit/dy2static/convert_call_func.py
index acbbe0e5d74f0..b5ce104d4dfe0 100644
--- a/python/paddle/jit/dy2static/convert_call_func.py
+++ b/python/paddle/jit/dy2static/convert_call_func.py
@@ -312,7 +312,7 @@ def dyfunc(x):
                 # Bound mothod will be convert into plain function after `convert_to_static`.
                 # So descriptor mechanism is used to bound `self` instance on function to
                 # keep it as bound method.
-                setattr(func, 'forward', forward_func.__get__(func))
+                func.forward = forward_func.__get__(func)
             except (IOError, OSError, TypeError):
                 # NOTE: func.forward may have been decorated.
                 func_self = None if func_self else func_self
diff --git a/python/paddle/jit/dy2static/program_translator.py b/python/paddle/jit/dy2static/program_translator.py
index 5b3eae6f8e3b6..9fbf2b6103538 100644
--- a/python/paddle/jit/dy2static/program_translator.py
+++ b/python/paddle/jit/dy2static/program_translator.py
@@ -314,8 +314,8 @@ def __init__(self, function, input_spec=None, **kwargs):
         # save the instance `self` while decorating a method of class.
 
         if inspect.ismethod(function):
-            self._dygraph_function = getattr(function, '__func__')
-            self._class_instance = getattr(function, '__self__')
+            self._dygraph_function = function.__func__
+            self._class_instance = function.__self__
 
             if not hasattr(self._class_instance, '_original_funcs'):
                 raise TypeError(
@@ -885,7 +885,7 @@ def __init__(self, func, class_instance, with_hook=False):
         self.need_apply_hook = (
             with_hook
             and isinstance(self.class_instance, layers.Layer)
-            and getattr(func, "__name__") == "forward"
+            and func.__name__ == "forward"
         )
 
     def apply_pre_hooks(self, inputs):
diff --git a/python/paddle/jit/dy2static/utils.py b/python/paddle/jit/dy2static/utils.py
index 0752e32b3fb10..28c0a0cfd1645 100644
--- a/python/paddle/jit/dy2static/utils.py
+++ b/python/paddle/jit/dy2static/utils.py
@@ -576,7 +576,7 @@ def func_prefix(func):
     # The 'forward' or 'another_forward' of 'TranslatedLayer' cannot be obtained
     # through 'func_name'. So set the special function name '__i_m_p_l__'.
     if hasattr(module, '__i_m_p_l__'):
-        callable_func = getattr(module, '__i_m_p_l__')
+        callable_func = module.__i_m_p_l__
         callable_func.__name__ = func_name
     elif hasattr(module, func_name):
         callable_func = getattr(module, func_name)
@@ -1120,11 +1120,11 @@ def __init__(self, root_node):
 
     def _reset_name_scope(self, node):
         # always reset the node as empty namescope.
-        setattr(node, "pd_scope", NameScope())
+        node.pd_scope = NameScope()
 
     def _get_name_scope(self, node):
         if not hasattr(node, "pd_scope"):
-            setattr(node, "pd_scope", NameScope())
+            node.pd_scope = NameScope()
         return node.pd_scope
 
     def _current_name_scope(self):
@@ -1224,11 +1224,7 @@ def post_func():
             )
 
         def pre_func():
-            setattr(
-                node,
-                "before_created",
-                self._nearest_function_scope().existed_vars(),
-            )
+            node.before_created = self._nearest_function_scope().existed_vars()
 
         self._visit_scope_node(node, pre_func, post_func)
 
diff --git a/python/paddle/nn/functional/vision.py b/python/paddle/nn/functional/vision.py
index 1acf1bcbc5f74..b7cebf4a58fbd 100644
--- a/python/paddle/nn/functional/vision.py
+++ b/python/paddle/nn/functional/vision.py
@@ -320,7 +320,7 @@ def grid_sample(
             'use_cudnn',
             use_cudnn,
         )
-        out = getattr(_legacy_C_ops, 'grid_sampler')(x, grid, *attrs)
+        out = _legacy_C_ops.grid_sampler(x, grid, *attrs)
     else:
         helper = LayerHelper("grid_sample", **locals())
         check_variable_and_dtype(x, 'x', ['float32', 'float64'], 'grid_sample')
diff --git a/python/paddle/nn/quant/qat/conv.py b/python/paddle/nn/quant/qat/conv.py
index 4c8e6915c15ad..f2ffc7b103ad7 100644
--- a/python/paddle/nn/quant/qat/conv.py
+++ b/python/paddle/nn/quant/qat/conv.py
@@ -30,18 +30,18 @@ def __init__(self, layer: Layer, q_config):
         super(QuantedConv2D, self).__init__()
 
         # For Conv2D
-        self._groups = getattr(layer, '_groups')
-        self._stride = getattr(layer, '_stride')
-        self._padding = getattr(layer, '_padding')
-        self._padding_mode = getattr(layer, '_padding_mode')
+        self._groups = layer._groups
+        self._stride = layer._stride
+        self._padding = layer._padding
+        self._padding_mode = layer._padding_mode
         if self._padding_mode != 'zeros':
-            self._reversed_padding_repeated_twice = getattr(
-                layer, '_reversed_padding_repeated_twice'
+            self._reversed_padding_repeated_twice = (
+                layer._reversed_padding_repeated_twice
             )
-        self._dilation = getattr(layer, '_dilation')
-        self._data_format = getattr(layer, '_data_format')
-        self.weight = getattr(layer, 'weight')
-        self.bias = getattr(layer, 'bias')
+        self._dilation = layer._dilation
+        self._data_format = layer._data_format
+        self.weight = layer.weight
+        self.bias = layer.bias
 
         self.weight_quanter = None
         self.activation_quanter = None
diff --git a/python/paddle/nn/quant/qat/linear.py b/python/paddle/nn/quant/qat/linear.py
index b089486531a1a..c0e015ce51c86 100644
--- a/python/paddle/nn/quant/qat/linear.py
+++ b/python/paddle/nn/quant/qat/linear.py
@@ -28,9 +28,9 @@ class QuantedLinear(ConvertibleQuantedLayer):
     def __init__(self, layer: Layer, q_config):
         super(QuantedLinear, self).__init__()
         # For Linear
-        self.weight = getattr(layer, 'weight')
-        self.bias = getattr(layer, 'bias')
-        self.name = getattr(layer, 'name')
+        self.weight = layer.weight
+        self.bias = layer.bias
+        self.name = layer.name
         # For FakeQuant
 
         self.weight_quanter = None
diff --git a/python/paddle/nn/quant/quant_layers.py b/python/paddle/nn/quant/quant_layers.py
index 257009a8ff1fb..9e7b4c55ba49e 100644
--- a/python/paddle/nn/quant/quant_layers.py
+++ b/python/paddle/nn/quant/quant_layers.py
@@ -533,18 +533,18 @@ def __init__(
     ):
         super().__init__()
         # For Conv2D
-        self._groups = getattr(layer, '_groups')
-        self._stride = getattr(layer, '_stride')
-        self._padding = getattr(layer, '_padding')
-        self._padding_mode = getattr(layer, '_padding_mode')
+        self._groups = layer._groups
+        self._stride = layer._stride
+        self._padding = layer._padding
+        self._padding_mode = layer._padding_mode
         if self._padding_mode != 'zeros':
-            self._reversed_padding_repeated_twice = getattr(
-                layer, '_reversed_padding_repeated_twice'
+            self._reversed_padding_repeated_twice = (
+                layer._reversed_padding_repeated_twice
             )
-        self._dilation = getattr(layer, '_dilation')
-        self._data_format = getattr(layer, '_data_format')
-        self.weight = getattr(layer, 'weight')
-        self.bias = getattr(layer, 'bias')
+        self._dilation = layer._dilation
+        self._data_format = layer._data_format
+        self.weight = layer.weight
+        self.bias = layer.bias
 
         # For FakeQuant
         self._conv2d_quant_axis = 0
@@ -654,14 +654,14 @@ def __init__(
         """
         super().__init__()
         # For Conv2DTranspose
-        self._groups = getattr(layer, '_groups')
-        self._stride = getattr(layer, '_stride')
-        self._padding = getattr(layer, '_padding')
-        self._output_padding = getattr(layer, 'output_padding')
-        self._dilation = getattr(layer, '_dilation')
-        self._data_format = getattr(layer, '_data_format')
-        self.weight = getattr(layer, 'weight')
-        self.bias = getattr(layer, 'bias')
+        self._groups = layer._groups
+        self._stride = layer._stride
+        self._padding = layer._padding
+        self._output_padding = layer.output_padding
+        self._dilation = layer._dilation
+        self._data_format = layer._data_format
+        self.weight = layer.weight
+        self.bias = layer.bias
         # For FakeQuant
         self._conv2d_transpose_quant_axis = 1
         if weight_quant_layer is not None:
@@ -748,9 +748,9 @@ def __init__(
     ):
         super().__init__()
         # For Linear
-        self.weight = getattr(layer, 'weight')
-        self.bias = getattr(layer, 'bias')
-        self.name = getattr(layer, 'name')
+        self.weight = layer.weight
+        self.bias = layer.bias
+        self.name = layer.name
         # For FakeQuant
         self._linear_quant_axis = 1
 
@@ -829,15 +829,15 @@ def __init__(
             act_quant_layer is None
         ), "When quantizing ColumnParallelLinear, act_quant_layer should be None."
 
-        self.weight = getattr(layer, 'weight')
-        self.bias = getattr(layer, 'bias')
-        self.name = getattr(layer, '_name')
+        self.weight = layer.weight
+        self.bias = layer.bias
+        self.name = layer._name
         # For FakeQuant
         self._linear_quant_axis = 1
 
-        self.is_mp = getattr(layer, 'is_mp')
-        self.model_parallel_group = getattr(layer, 'model_parallel_group')
-        self.gather_output = getattr(layer, 'gather_output')
+        self.is_mp = layer.is_mp
+        self.model_parallel_group = layer.model_parallel_group
+        self.gather_output = layer.gather_output
 
         self._fake_quant_weight = _get_fake_quant_type(
             weight_quantize_type,
@@ -923,15 +923,15 @@ def __init__(
         ), "When quantizing RowParallelLinear, act_quant_layer cannot defined by yourself."
 
         # For Linear
-        self.weight = getattr(layer, 'weight')
-        self.bias = getattr(layer, 'bias')
-        self.name = getattr(layer, '_name')
+        self.weight = layer.weight
+        self.bias = layer.bias
+        self.name = layer._name
         # For FakeQuant
         self._linear_quant_axis = 1
 
-        self.input_is_parallel = getattr(layer, 'input_is_parallel')
-        self.is_mp = getattr(layer, 'is_mp')
-        self.model_parallel_group = getattr(layer, 'model_parallel_group')
+        self.input_is_parallel = layer.input_is_parallel
+        self.is_mp = layer.is_mp
+        self.model_parallel_group = layer.model_parallel_group
 
         self._fake_quant_weight = _get_fake_quant_type(
             weight_quantize_type,