diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/gkdropout.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/gkdropout.py
index 8101c6c7..74870291 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/gkdropout.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/gkdropout.py
@@ -22,12 +22,27 @@ class GkDropout(Expander):
     """GkDropout expander"""
 
     def _expand(self, graph_builder):
+        """
+        对输入数据进行dropout操作。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            tuple: 包含两个元素，第一个是执行dropout操作后的结果，第二个是生成的掩码。
+
+        """
+        # 获取输入数据和掩码
         input_x, input_mask = self.inputs
+        # 获取保持概率
         keep_prob = self.attrs['keep_prob']
 
+        # 计算保持概率的倒数
         r_keep_prob = graph_builder.value(input_x.dtype, 1.0 / keep_prob)
+        # 计算保持概率
         keep_prob = graph_builder.value(input_x.dtype, keep_prob)
 
+        # 如果掩码的数据类型与输入数据类型不同，则进行类型转换
         if input_mask.dtype != input_x.dtype:
             input_mask = graph_builder.emit('Cast', [input_mask], attrs={'dst_type': input_x.dtype})
         mask = graph_builder.emit('LessEqual', [input_mask, keep_prob])  # output is bool type
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/identity.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/identity.py
index fe500660..d0abf4c3 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/identity.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/identity.py
@@ -20,6 +20,16 @@ class Identity(Expander):
     """Identity expander"""
 
     def _expand(self, graph_builder):
-        input_x = self.inputs[0]
-        result = graph_builder.emit('Reshape', [input_x], attrs={'shape': input_x.shape})
-        return result
+        """
+        对输入数据进行重塑操作。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于构建计算图。
+
+        Returns:
+            Tensor: 重塑后的输入数据。
+
+        """
+        input_x = self.inputs[0] #  获取输入数据
+        result = graph_builder.emit('Reshape', [input_x], attrs={'shape': input_x.shape}) #  使用图构建器对象构建计算图，对输入数据进行重塑操作
+        return result #  返回重塑后的输入数据
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/layernorm.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/layernorm.py
index eaa44140..791445c1 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/layernorm.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/layernorm.py
@@ -25,67 +25,107 @@ class LayerNorm(Expander):
     """LayerNorm expander"""
 
     def _expand(self, graph_builder):
+        """
+        对输入进行扩展处理，包括批量归一化操作。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于构建计算图。
+
+        Returns:
+            tuple: 包含三个元素的元组，分别是处理后的输入、均值和方差。
+                - res (Tensor): 处理后的输入张量。
+                - mean (Tensor): 输入的均值张量。
+                - variance (Tensor): 输入的方差张量。
+
+        """
+        # 获取输入数据
         input_x, input_gamma, input_beta = self.inputs
+        # 获取处理器类型
         processor = self.processor
+        # 获取归一化开始轴
         begin_norm_axis = self.attrs['begin_norm_axis']
+        # 获取epsilon值
         epsilon = self.attrs['epsilon']
 
+        # 获取输入数据的原始数据类型
         ori_dtype = input_x.dtype
+        # 如果处理器类型为aicore且输入数据类型为float16，则将输入数据类型转换为float32
         if processor == 'aicore' and ori_dtype == 'float16':
             input_x = graph_builder.emit('Cast', [input_x], attrs={'dst_type': 'float32'})
             input_gamma = graph_builder.emit('Cast', [input_gamma], attrs={'dst_type': 'float32'})
             input_beta = graph_builder.emit('Cast', [input_beta], attrs={'dst_type': 'float32'})
 
+        # 获取输入数据的原始形状
         ori_shape_x = input_x.shape
+        # 如果输入数据的格式为FRAC_NZ，则根据FRAC_NZ格式获取输入数据的形状
         if input_x.data_format == DF.FRAC_NZ:
             ori_shape_x = infer_shape_from_fractalnz(input_x.shape)
 
         # Calculate the scaling ratio of the average
+        # 如果begin_norm_axis小于0，则将其加上ori_shape_x的长度
         if begin_norm_axis < 0:
             begin_norm_axis += len(ori_shape_x)
 
+        # 定义reduce_axis，用于存储需要归一化的维度
         reduce_axis = ()
+        # 遍历ori_shape_x，如果维度大于begin_norm_axis或者等于begin_norm_axis，则将其加入reduce_axis
         for i, _ in enumerate(ori_shape_x):
             if i > begin_norm_axis or i == begin_norm_axis:
                 reduce_axis = reduce_axis + (i,)
 
+        # 计算reduce_elts，即需要归一化的维度上的元素个数
         reduce_elts = 1.0
         for i in reduce_axis:
             reduce_elts *= ori_shape_x[i]
         # after reduced
+        # 获取归一化后的ori_shape_x
         ori_reduced_shape_x = get_reduced_ori_shape(ori_shape_x, reduce_axis)
 
+        # 定义axis，用于存储归一化的维度
         axis = reduce_axis
+        # 如果input_x的数据格式为DF.FRAC_NZ，则将axis转换为frac_z轴
         if input_x.data_format == DF.FRAC_NZ:
             axis = to_frac_z_axis(ori_shape_x, reduce_axis)
 
+        # 计算mean_cof_v，即归一化系数
         mean_cof_v = graph_builder.value(input_x.dtype, 1.0 / reduce_elts)
 
         # Calculate mean
+        # 计算输入张量的均值
         mean_red = graph_builder.emit('ReduceSum', [input_x], attrs={'reduce_axis': axis, 'keep_dims': True})
+        # 将均值乘以系数
         mean = graph_builder.emit('Mul', [mean_red, mean_cof_v])
+        # 如果输入张量的数据格式为DF.FRAC_NZ，则对均值进行重整
         if input_x.data_format == DF.FRAC_NZ:
             mean = graph_builder.emit('Reshape', [mean], attrs={'shape': ori_reduced_shape_x})
 
         # Calculate variance
-        variance_sub = graph_builder.emit('Sub', [input_x, mean])
-        variance_mul = graph_builder.emit('Mul', [variance_sub, variance_sub])
-        variance_red = graph_builder.emit('ReduceSum', [variance_mul], attrs={'reduce_axis': axis, 'keep_dims': True})
-        variance = graph_builder.emit('Mul', [variance_red, mean_cof_v])
+        # 计算方差
+        variance_sub = graph_builder.emit('Sub', [input_x, mean])  # 计算输入与均值的差值
+        variance_mul = graph_builder.emit('Mul', [variance_sub, variance_sub])  # 计算差值的平方
+        variance_red = graph_builder.emit('ReduceSum', [variance_mul], attrs={'reduce_axis': axis, 'keep_dims': True})  # 对差值的平方求和
+        variance = graph_builder.emit('Mul', [variance_red, mean_cof_v])  # 计算方差
+        # 如果输入数据的格式为DF.FRAC_NZ，则对方差进行reshape操作
         if input_x.data_format == DF.FRAC_NZ:
             variance = graph_builder.emit('Reshape', [variance], attrs={'shape': ori_reduced_shape_x})
 
         # Calculate normalize
+        # 计算输入x与均值之间的差值
         normalize_sub = graph_builder.emit('Sub', [input_x, mean])
+        # 创建一个epsilon值，用于防止除零错误
         epsilon_v = graph_builder.value(input_x.dtype, epsilon)
+        # 计算方差加上epsilon的值
         normalize_add = graph_builder.emit('Add', [variance, epsilon_v])
         normlize_rsqrt = graph_builder.emit('Rsqrt', [normalize_add])
         normalize_mul = graph_builder.emit('Mul', [normalize_sub, normlize_rsqrt])
 
         # Calculate scale and translate
+        # 计算归一化后的乘积
         scale_mul = graph_builder.emit('Mul', [normalize_mul, input_gamma])
+        # 计算最终结果
         res = graph_builder.emit('Add', [scale_mul, input_beta])
 
+        # 如果处理器为aicore且原始数据类型为float16，则将结果、均值和方差转换为float16
         if processor == 'aicore' and ori_dtype == 'float16':
             res = graph_builder.emit('Cast', [res], attrs={'dst_type': 'float16'})
             mean = graph_builder.emit('Cast', [mean], attrs={'dst_type': 'float16'})
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/layernorm_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/layernorm_grad.py
index 2ae7078b..4facbef1 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/layernorm_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/layernorm_grad.py
@@ -23,13 +23,33 @@ class LayerNormGrad(Expander):
     """LayerNormGrad expander"""
 
     def _expand(self, graph_builder):
+        """
+        对输入进行扩展操作。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            tuple: 包含dx, dg, db的元组。
+                dx (Tensor): 梯度相对于输入x的导数。
+                dg (Tensor): 梯度相对于gamma的导数。
+                db (Tensor): 梯度相对于beta的导数。
+
+        """
+        # 获取输入参数
         x, dy, variance, mean, gamma = self.inputs
+        # 获取处理器类型
         processor = self.processor
+        # 获取归一化轴的起始位置
         begin_norm_axis = self.attrs['begin_norm_axis']
+        # 获取参数轴的起始位置
         begin_params_axis = self.attrs['begin_params_axis']
+        # 获取epsilon值，默认为1e-12
         epsilon = self.attrs['epsilon'] if 'epsilon' in self.attrs else 1e-12
 
+        # 获取输入数据的原始数据类型
         ori_dtype = x.dtype
+        # 如果处理器类型为aicore且数据类型为float16，则将输入数据转换为float32
         if processor == 'aicore' and ori_dtype == 'float16':
             x = graph_builder.emit('Cast', [x], attrs={'dst_type': 'float32'})
             dy = graph_builder.emit('Cast', [dy], attrs={'dst_type': 'float32'})
@@ -37,77 +57,121 @@ class LayerNormGrad(Expander):
             mean = graph_builder.emit('Cast', [mean], attrs={'dst_type': 'float32'})
             gamma = graph_builder.emit('Cast', [gamma], attrs={'dst_type': 'float32'})
 
+        # 如果归一化轴的起始位置小于0，则将其转换为正数
         if begin_norm_axis < 0:
             begin_norm_axis += len(x.shape)
+        # 如果参数轴的起始位置小于0，则将其转换为正数
         if begin_params_axis < 0:
             begin_params_axis += len(x.shape)
 
+        # 获取归一化轴和参数轴的范围
         norm_axis = tuple(range(begin_norm_axis, len(x.shape)))
         param_axis = tuple(range(0, begin_params_axis))
 
+        # 计算归一化轴的维度乘积
         reduce_size = 1.0
         for i in norm_axis:
             reduce_size *= x.shape[i]
 
         # set some constant val.
+        # 计算epsilon的值
         eps = graph_builder.value(x.dtype, epsilon)
+        # 计算-0.5的值
         const_neg_half = graph_builder.value(x.dtype, -0.5)
+        # 计算-2.0的值
         const_neg_two = graph_builder.value(x.dtype, -2.0)
+        # 计算2.0的值
         const_two = graph_builder.value(x.dtype, 2.0)
+        # 计算-1.0的值
         const_neg_one = graph_builder.value(x.dtype, -1.0)
+        # 计算mean_cof的值
         mean_cof = graph_builder.value(x.dtype, (1.0 / reduce_size))
 
         # cal dg db
+        # 计算方差和eps的和
         var_eps = graph_builder.emit('Add', [variance, eps])
+        # 计算方差和eps的和的对数
         var_eps_log = graph_builder.emit('Log', [var_eps])
+        # 计算方差和eps的对数乘以-0.5
         var_eps_mul = graph_builder.emit('Mul', [var_eps_log, const_neg_half])
+        # 计算方差和eps的对数乘以-0.5的指数
         rsqrt_var_eps = graph_builder.emit('Exp', [var_eps_mul])
 
+        # 计算x和mean的差
+        # 计算输入x减去均值
         x_sub_mean = graph_builder.emit('Sub', [x, mean])
 
+        # 计算x减去均值乘以rsqrt_var_eps
         x_sub_mean_mul_rsqrt_var_eps = graph_builder.emit('Mul', [rsqrt_var_eps, x_sub_mean])
+        # 计算dy乘以x减去均值乘以rsqrt_var_eps
         dg_mul = graph_builder.emit('Mul', [dy, x_sub_mean_mul_rsqrt_var_eps])
+        # 计算dg，对dg_mul进行求和，reduce_axis为param_axis，keep_dims为False
         dg = graph_builder.emit('ReduceSum', [dg_mul], attrs={'reduce_axis': param_axis, 'keep_dims': False})
+        # 计算db，对dy进行求和，reduce_axis为param_axis，keep_dims为False
         db = graph_builder.emit('ReduceSum', [dy], attrs={'reduce_axis': param_axis, 'keep_dims': False})
 
         # pd_var
+        # 计算tmp_var_eps
         tmp_var_eps = graph_builder.emit('Mul', [rsqrt_var_eps, rsqrt_var_eps])
+        # 计算r_tmp_var_eps
         r_tmp_var_eps = graph_builder.emit('Mul', [rsqrt_var_eps, tmp_var_eps])
 
+        # 计算dy_mul_gamma
         dy_mul_gamma = graph_builder.emit('Mul', [dy, gamma])
+        # 计算tmp_mul
         tmp_mul = graph_builder.emit('Mul', [dy_mul_gamma, x_sub_mean])
+        # 计算padvar_mul1
         padvar_mul1 = graph_builder.emit('ReduceSum', [tmp_mul], attrs={'reduce_axis': norm_axis, 'keep_dims': True})
+        # 计算padvar_mul3
         padvar_mul3 = graph_builder.emit('Mul', [padvar_mul1, r_tmp_var_eps])
+        # 计算pd_var
         pd_var = graph_builder.emit('Mul', [padvar_mul3, const_neg_half])
 
         # pd_mean
+        # 计算pdmean1_sum，使用ReduceSum函数，输入为dy_mul_gamma，归约轴为norm_axis，保持维度为True
         pdmean1_sum = graph_builder.emit('ReduceSum', [dy_mul_gamma],
                                          attrs={'reduce_axis': norm_axis, 'keep_dims': True})
+        # 计算neg_rsqrt_var_eps，使用Mul函数，输入为rsqrt_var_eps和const_neg_one
         neg_rsqrt_var_eps = graph_builder.emit('Mul', [rsqrt_var_eps, const_neg_one])
+        # 计算pd_mean_1，使用Mul函数，输入为neg_rsqrt_var_eps和pdmean1_sum
         pd_mean_1 = graph_builder.emit('Mul', [neg_rsqrt_var_eps, pdmean1_sum])
 
+        # 计算pdmean2_mul1，使用Mul函数，输入为const_neg_two和x_sub_mean
         pdmean2_mul1 = graph_builder.emit('Mul', [const_neg_two, x_sub_mean])
+        # 计算pdmean2_sum，使用ReduceSum函数，输入为pdmean2_mul1，归约轴为norm_axis，保持维度为True
         pdmean2_sum = graph_builder.emit('ReduceSum', [pdmean2_mul1],
                                          attrs={'reduce_axis': norm_axis, 'keep_dims': True})
+        # 计算pdmean2_mul3，使用Mul函数，输入为pdmean2_sum和mean_cof
         pdmean2_mul3 = graph_builder.emit('Mul', [pdmean2_sum, mean_cof])
+        # 计算pd_mean_2，使用Mul函数，输入为pdmean2_mul3和pd_var
         pd_mean_2 = graph_builder.emit('Mul', [pdmean2_mul3, pd_var])
 
+        # 计算pd_mean，使用Add函数，输入为pd_mean_1和pd_mean_2
         pd_mean = graph_builder.emit('Add', [pd_mean_1, pd_mean_2])
 
         # cal dx
+        # 计算pd_x_1
         pd_x_1 = graph_builder.emit('Mul', [dy_mul_gamma, rsqrt_var_eps])
 
+        # 计算pdx2_mul
         pdx2_mul = graph_builder.emit('Mul', [pd_var, x_sub_mean])
+        # 计算pdx2_mul_two
         pdx2_mul_two = graph_builder.emit('Mul', [pdx2_mul, const_two])
+        # 计算pd_x_2
         pd_x_2 = graph_builder.emit('Mul', [pdx2_mul_two, mean_cof])
 
+        # 计算pd_x_3
         pd_x_3 = graph_builder.emit('Mul', [pd_mean, mean_cof])
 
+        # 计算dx_tmp
         dx_tmp = graph_builder.emit('Add', [pd_x_1, pd_x_2])
+        # 计算dx
         dx = graph_builder.emit('Add', [dx_tmp, pd_x_3])
 
+        # 如果处理器为aicore且原始数据类型为float16，则将dx、dg、db转换为float16
         if processor == 'aicore' and ori_dtype == 'float16':
             dx = graph_builder.emit('Cast', [dx], attrs={'dst_type': 'float16'})
             dg = graph_builder.emit('Cast', [dg], attrs={'dst_type': 'float16'})
             db = graph_builder.emit('Cast', [db], attrs={'dst_type': 'float16'})
+        # 返回dx、dg、db
         return dx, dg, db
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/logsoftmax.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/logsoftmax.py
index 27aa8035..691f4780 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/logsoftmax.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/logsoftmax.py
@@ -23,24 +23,49 @@ class LogSoftmax(Expander):
     """LogSoftmax expander"""
 
     def _expand(self, graph_builder):
+        """
+        对输入数据进行Softmax归一化。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于构建计算图。
+
+        Returns:
+            Tensor: Softmax归一化后的结果。
+
+        """
+        # 获取输入数据
         input_x = self.inputs[0]
+        # 获取轴参数
         axis = self.attrs['axis']
+        # 获取处理器类型
         processor = self.processor
 
+        # 如果轴参数是整数，则将其转换为元组
         if isinstance(axis, int):
             axis = (axis,)
 
+        # 获取输入数据的原始数据类型
         ori_dtype = input_x.dtype
+        # 如果原始数据类型不是float16且处理器类型是aicore，则将输入数据转换为float16
         if ori_dtype != "float16" and processor == "aicore":
             input_x_f16 = graph_builder.emit('Cast', [input_x], attrs={'dst_type': 'float16'})
+            # 对转换后的数据进行ReduceMax操作
             max_x_f16 = graph_builder.emit('ReduceMax', [input_x_f16], attrs={'reduce_axis': axis, 'keep_dims': True})
+            # 将ReduceMax操作的结果转换回原始数据类型
             max_x = graph_builder.emit('Cast', [max_x_f16], attrs={'dst_type': ori_dtype})
         else:
+            # 对输入数据进行ReduceMax操作
             max_x = graph_builder.emit('ReduceMax', [input_x], attrs={'reduce_axis': axis, 'keep_dims': True})
+        # 计算输入数据与ReduceMax操作结果的差值
         data_sub = graph_builder.emit('Sub', [input_x, max_x])
+        # 计算差值的指数
         data_exp = graph_builder.emit('Exp', [data_sub])
+        # 对指数结果进行ReduceSum操作
         data_expsum = graph_builder.emit('ReduceSum', [data_exp], attrs={'reduce_axis': axis, 'keep_dims': True})
+        # 计算ReduceSum结果的log
         log_expsum = graph_builder.emit('Log', [data_expsum])
+        # 计算差值与log的差值
         result = graph_builder.emit('Sub', [data_sub, log_expsum])
 
+        # 返回结果
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/logsoftmax_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/logsoftmax_grad.py
index 9a075d7c..7af59ed0 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/logsoftmax_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/logsoftmax_grad.py
@@ -23,14 +23,32 @@ class LogSoftmaxGrad(Expander):
     """LogSoftmaxGrad expander"""
 
     def _expand(self, graph_builder):
+        """
+        对输入进行扩展操作。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            Tensor: 扩展操作的结果。
+
+        """
+        # 获取输入的logits和dy
         input_logits, input_dy = self.inputs
+        # 获取axis参数
         axis = self.attrs['axis']
+        # 如果axis是整数，则将其转换为元组
         if isinstance(axis, int):
             axis = (axis,)
 
+        # 计算softmax
         softmax = graph_builder.emit('Exp', [input_logits])
+        # 计算dy的sum
         dy_sum = graph_builder.emit('ReduceSum', [input_dy], attrs={'reduce_axis': axis, 'keep_dims': True})
+        # 计算softmax和dy_sum的乘积
         mul_result = graph_builder.emit('Mul', [softmax, dy_sum])
+        # 计算input_dy和mul_result的差
         result = graph_builder.emit('Sub', [input_dy, mul_result])
 
+        # 返回结果
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/matmul.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/matmul.py
index 8af69e0b..1d888f58 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/matmul.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/matmul.py
@@ -25,48 +25,139 @@ class MatMul(Expander):
     """
 
     def __init__(self, expand_info):
+        """
+        初始化MatMul类实例。
+
+        Args:
+            expand_info (dict): 扩展信息字典，包含操作所需的额外信息。
+
+        Attributes:
+            transpose_a (bool): 是否对矩阵A进行转置。
+            transpose_b (bool): 是否对矩阵B进行转置。
+            left_format (str): 矩阵A的数据格式。
+            right_format (str): 矩阵B的数据格式。
+            shape_a (tuple): 矩阵A的形状。
+            shape_b (tuple): 矩阵B的形状。
+
+        """
+        # 调用父类的初始化方法
         super(MatMul, self).__init__(expand_info)
+        # 获取transpose_a属性
         self.transpose_a = self.attrs['transpose_a']
+        # 获取transpose_b属性
         self.transpose_b = self.attrs['transpose_b']
+        # 获取left_format属性
         self.left_format = self.attrs['left_format']
+        # 获取right_format属性
         self.right_format = self.attrs['right_format']
+        # 获取输入A的shape
         self.shape_a = self.inputs[0]['shape']
+        # 获取输入B的shape
         self.shape_b = self.inputs[1]['shape']
 
     def _optimize_to_mul(self):
+        """
+        检查是否可以用乘法（mul）替换矩阵乘法（matmul）
+
+        Args:
+            无
+
+        Returns:
+            bool: 如果可以用乘法替换矩阵乘法，返回True；否则返回False。
+
+        """
         """check if matmul can be replace by mul"""
+        # 如果处理器不是'aicore'，或者左格式或右格式不是默认格式，则返回False
         if self.processor != 'aicore' or self.left_format != DF.DEFAULT or self.right_format != DF.DEFAULT:
             return False
+        # 如果transpose_a为True，则k_a为shape_a的倒数第二个维度，否则为shape_a的倒数第一个维度
         k_a = self.shape_a[-2] if self.transpose_a else self.shape_a[-1]
+        # 如果transpose_b为True，则k_b为shape_b的倒数第一个维度，否则为shape_b的倒数第二个维度
         k_b = self.shape_b[-1] if self.transpose_b else self.shape_b[-2]
+        # 如果k_a或k_b不等于1，则返回False
         if k_a != 1 or k_b != 1:
             return False
+        # 否则返回True
         return True
 
     def _check(self):
+        """
+        检查输入个数是否满足矩阵乘法的要求。
+
+        Args:
+            无
+
+        Returns:
+            无
+
+        Raises:
+            GKException: 如果输入的个数小于2，则抛出GKException异常，提示信息为 "For 'MatMul', inputs number should bigger than 1, but got {}."，其中{}为输入的个数。
+
+        """
+        # 获取输入的个数
         input_num = len(self.inputs)
+        # 如果输入的个数小于2，抛出异常
         if input_num < 2:
             raise GKException("For 'MatMul', inputs number should bigger than 1, but got {}.".format(input_num))
 
     def _expand(self, graph_builder):
+        """
+        将MatMul或BatchMatMul操作替换为Mul操作。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            Node: Mul操作的结果节点。
+
+        Raises:
+            GKException: 如果不需要将MatMul/BatchMatMul替换为Mul操作，则引发异常。
+
+        """
+        # 定义一个函数，用于转置shape
         def transpose(shape):
+            """
+            将给定的shape进行转置操作。
+
+            Args:
+                shape (tuple): 输入的shape，为一个元组，表示多维数组的形状。
+
+            Returns:
+                list: 转置后的shape，以列表形式返回。
+
+            """
+            # 将shape转换为列表
             trans_shape = list(shape)
+            # 将shape的倒数第二个元素和倒数第一个元素交换位置
             trans_shape[-2] = shape[-1]
             trans_shape[-1] = shape[-2]
+            # 返回转置后的shape
             return trans_shape
+        # 如果不需要优化为乘法，则抛出异常
         if not self._optimize_to_mul():
             raise GKException("MatMul/BatchMatMul do not need to be replaced by Mul")
         # Matmul is replaced by Mul([b m k], [b k n]) when k==1
+        # 获取输入a
         input_a = self.inputs[0]
+        # 获取输入b
         input_b = self.inputs[1]
+        # 如果transpose_a为True，则对输入a进行转置
         if self.transpose_a:
+            # 获取输入a的转置形状
             shape_a_trans = transpose(self.shape_a)
+            # 对输入a进行转置
             input_a = graph_builder.emit('Reshape', [input_a], attrs={'shape': shape_a_trans})
+        # 如果transpose_b为True，则对输入b进行转置
         if self.transpose_b:
+            # 获取输入b的转置形状
             shape_b_trans = transpose(self.shape_b)
+            # 对输入b进行转置
             input_b = graph_builder.emit('Reshape', [input_b], attrs={'shape': shape_b_trans})
+        # 对输入a和输入b进行乘法运算
         result = graph_builder.emit('Mul', [input_a, input_b])
+        # 如果dst_type在attrs中，并且输入a的数据类型与dst_type不同，则对结果进行类型转换
         if 'dst_type' in self.attrs and self.inputs[0].dtype != self.attrs['dst_type']:
+            # 对结果进行类型转换
             result = graph_builder.emit('Cast', [result], attrs={'dst_type': self.attrs['dst_type']})
         return result
 
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/maximum_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/maximum_grad.py
index d29989cb..0e62e7b7 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/maximum_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/maximum_grad.py
@@ -23,35 +23,76 @@ class MaximumGrad(Expander):
     """MaximumGrad expander"""
 
     def _check(self):
+        """
+        检查MaximumGrad的属性是否符合要求。
+
+        Args:
+            无
+
+        Returns:
+            返回父类的检查结果。
+
+        Raises:
+            GKException: 当 'grad_x' 和 'grad_y' 的值都为 False 时抛出异常。
+
+        """
+        # 如果attr 'grad_x'和'grad_y'的值都为False，则抛出异常
         if not self.attrs.get('grad_x', True) and not self.attrs.get('grad_y', True):
             raise GKException("For 'MaximumGrad', value of attr 'grad_x' and 'grad_y' should be False, but got {} and "
                               "{}".format(self.attrs.get('grad_x'), self.attrs.get('grad_y')))
+        # 调用父类的方法
         return super()._check()
 
     def _expand(self, graph_builder):
+        """
+        根据输入计算梯度，并返回两个梯度结果。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于构建计算图。
+
+        Returns:
+            tuple: 包含两个梯度结果的元组，第一个元素为对输入x的梯度，第二个元素为对输入y的梯度。
+
+        """
+        # 获取输入的x、y和dout
         input_x, input_y, input_dout = self.inputs
+        # 比较x和y的大小，返回一个布尔值
         ge_result = graph_builder.emit('GreaterEqual', [input_x, input_y])
+        # 将布尔值转换为与x相同的类型
         ge_result = graph_builder.emit('Cast', [ge_result], attrs={'dst_type': input_x.dtype})
+        # 计算dx，即x的梯度
         dx = graph_builder.emit('Mul', [ge_result, input_dout])
+        # 计算dy，即y的梯度
         dy = graph_builder.emit('Sub', [input_dout, dx])
 
+        # 获取dx和dy的reduce轴
         reduce_axis_x = MinimumGrad.get_reduce_axis(input_x.shape, dx.shape)
         reduce_axis_y = MinimumGrad.get_reduce_axis(input_y.shape, dy.shape)
+        # 如果dx有reduce轴
         if reduce_axis_x:
+            # 对dx进行求和
             dx_reduce = graph_builder.emit('ReduceSum', [dx], attrs={'reduce_axis': reduce_axis_x, 'keep_dims': False})
+            # 如果dx_reduce的形状与input_x的形状不同，则进行reshape
             if dx_reduce.shape != input_x.shape:
                 dx_out = graph_builder.emit('Reshape', [dx_reduce], attrs={'shape': input_x.shape})
+            # 否则，dx_out等于dx_reduce
             else:
                 dx_out = dx_reduce
+        # 否则，dx_out等于dx
         else:
             dx_out = dx
 
+        # 如果dy有reduce轴
         if reduce_axis_y:
+            # 对dy进行求和
             dy_reduce = graph_builder.emit('ReduceSum', [dy], attrs={'reduce_axis': reduce_axis_y, 'keep_dims': False})
+            # 如果dy_reduce的形状与input_y的形状不同，则进行reshape
             if dy_reduce.shape != input_y.shape:
                 dy_out = graph_builder.emit('Reshape', [dy_reduce], attrs={'shape': input_y.shape})
+            # 否则，dy_out等于dy_reduce
             else:
                 dy_out = dy_reduce
+        # 否则，dy_out等于dy
         else:
             dy_out = dy
 
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/minimum_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/minimum_grad.py
index 8772fb0e..0a98d1df 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/minimum_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/minimum_grad.py
@@ -22,59 +22,117 @@ class MinimumGrad(Expander):
     """MinimumGrad expander"""
 
     def _check(self):
+        """
+        检查MinimumGrad类的属性是否满足要求。
+
+        Args:
+            无
+
+        Returns:
+            bool: 如果属性符合要求，则返回True，否则抛出GKException异常。
+
+        Raises:
+            GKException: 如果MinimumGrad类的属性'grad_x'和'grad_y'均为False，则抛出此异常。
+
+        """
+        # 如果attr 'grad_x'和'grad_y'的值都为False，则抛出异常
         if not self.attrs.get('grad_x', True) and not self.attrs.get('grad_y', True):
             raise GKException("For 'MinimumGrad', value of attr 'grad_x' and 'grad_y' should be False, but got {} and "
                               "{}".format(self.attrs.get('grad_x'), self.attrs.get('grad_y')))
+        # 调用父类的方法
         return super(MinimumGrad, self)._check()
 
     def _expand(self, graph_builder):
+        """
+        计算两个输入的梯度。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于在图中执行操作。
+
+        Returns:
+            tuple: 包含两个梯度结果的元组。
+
+        """
+        # 输入参数
         input_x, input_y, input_dout = self.inputs
 
-        le_result = graph_builder.emit('LessEqual', [input_x, input_y])
-        le_result = graph_builder.emit('Cast', [le_result], attrs={'dst_type': input_x.dtype})
-        dx = graph_builder.emit('Mul', [le_result, input_dout])
-        dy = graph_builder.emit('Sub', [input_dout, dx])
+        # 执行 LessEqual 操作
+        le_result = graph_builder.emit('LessEqual', [input_x, input_y])  # 执行 LessEqual 操作
+        # 将结果转换为与 input_x 相同的数据类型
+        le_result = graph_builder.emit('Cast', [le_result], attrs={'dst_type': input_x.dtype})  # 将结果转换为与 input_x 相同的数据类型
+        # 执行 Mul 操作，将 le_result 和 input_dout 相乘
+        dx = graph_builder.emit('Mul', [le_result, input_dout])  # 执行 Mul 操作，将 le_result 和 input_dout 相乘
+        # 执行 Sub 操作，用 input_dout 减去 dx
+        dy = graph_builder.emit('Sub', [input_dout, dx])  # 执行 Sub 操作，用 input_dout 减去 dx
 
+        # 对于 minimumgrad 操作，输出形状应与输入形状相同，
+        # 但某些元素级操作可能会广播输入形状，
+        # 导致输出形状不等于原始输入形状，因此需要减少输出来使它们相等
         # for minimumgrad op, output_shape should be equal to input_shape,
         # but some elementwise operating may broadcast input_shape
         # then output_shape not equal to original input_shape, so need to reduce output to let them equal
-        reduce_axis_x = self.get_reduce_axis(input_x.shape, dx.shape)
-        reduce_axis_y = self.get_reduce_axis(input_y.shape, dy.shape)
+        reduce_axis_x = self.get_reduce_axis(input_x.shape, dx.shape)  # 获取 x 的减少轴
+        reduce_axis_y = self.get_reduce_axis(input_y.shape, dy.shape)  # 获取 y 的减少轴
         if reduce_axis_x:
-            dx_reduce = graph_builder.emit('ReduceSum', [dx], attrs={'reduce_axis': reduce_axis_x, 'keep_dims': False})
+            # 如果存在减少轴，执行 ReduceSum 操作
+            dx_reduce = graph_builder.emit('ReduceSum', [dx], attrs={'reduce_axis': reduce_axis_x, 'keep_dims': False})  # 执行 ReduceSum 操作
             if dx_reduce.shape != input_x.shape:
-                dx_out = graph_builder.emit('Reshape', [dx_reduce], attrs={'shape': input_x.shape})
+                # 如果减少后的形状不等于原始输入形状，则进行 Reshape 操作
+                dx_out = graph_builder.emit('Reshape', [dx_reduce], attrs={'shape': input_x.shape})  # 如果减少后的形状不等于原始输入形状，则进行 Reshape 操作
             else:
-                dx_out = dx_reduce
+                dx_out = dx_reduce  # 否则直接使用减少后的结果
         else:
-            dx_out = dx
+            dx_out = dx  # 如果没有减少轴，则直接使用 dx
 
         if reduce_axis_y:
-            dy_reduce = graph_builder.emit('ReduceSum', [dy], attrs={'reduce_axis': reduce_axis_y, 'keep_dims': False})
+            # 如果存在减少轴，执行 ReduceSum 操作
+            dy_reduce = graph_builder.emit('ReduceSum', [dy], attrs={'reduce_axis': reduce_axis_y, 'keep_dims': False})  # 执行 ReduceSum 操作
             if dy_reduce.shape != input_y.shape:
-                dy_out = graph_builder.emit('Reshape', [dy_reduce], attrs={'shape': input_y.shape})
+                # 如果减少后的形状不等于原始输入形状，则进行 Reshape 操作
+                dy_out = graph_builder.emit('Reshape', [dy_reduce], attrs={'shape': input_y.shape})  # 如果减少后的形状不等于原始输入形状，则进行 Reshape 操作
             else:
-                dy_out = dy_reduce
+                dy_out = dy_reduce  # 否则直接使用减少后的结果
         else:
-            dy_out = dy
+            dy_out = dy  # 如果没有减少轴，则直接使用 dy
 
-        # output two results, regardless of grad_x and grad_y
+        # 输出两个结果，
         return dx_out, dy_out
 
     @staticmethod
     def get_reduce_axis(original_shape, broadcast_shape):
+        """
+        计算最终输出形状的归约轴。
+
+        Args:
+            original_shape (tuple of int): 原始形状，一个包含整数的元组。
+            broadcast_shape (tuple of int): 广播形状，一个包含整数的元组。
+
+        Returns:
+            list of int: 归约轴列表，表示在最终输出形状中需要归约的轴索引。
+
+        Raises:
+            ValueError: 如果original_shape的长度大于broadcast_shape的长度，或者original_shape和broadcast_shape无法广播。
+
+        """
         """compute reduce axis for final output_shape"""
+        # 如果original_shape的长度大于broadcast_shape的长度
         if len(original_shape) > len(broadcast_shape):
             raise ValueError("For 'MinimumGrad', the length of original_shape should be less than or equal to the "
                              "length of broadcast_shape, but got {} and {}".format(original_shape, broadcast_shape))
 
+        # 创建一个tmp_shape列表，长度为broadcast_shape的长度，前面填充1，后面填充original_shape的元素
         tmp_shape = [1] * (len(broadcast_shape) - len(original_shape)) + original_shape
         reduce_axis = []
+        # 遍历tmp_shape中的每个元素
         for idx, _ in enumerate(tmp_shape):
+            # 如果tmp_shape中的元素与broadcast_shape中的对应元素不相等
             if tmp_shape[idx] != broadcast_shape[idx]:
+                # 如果tmp_shape中的元素为1
                 if tmp_shape[idx] == 1:
+                    # 将当前索引添加到reduce_axis列表中
                     reduce_axis.append(idx)
                 else:
+                    # 抛出异常，表示original_shape和broadcast_shape无法广播
                     raise ValueError("For 'MinimumGrad', original_shape {} and broadcast_shape {} can not broadcast."
                                      .format(original_shape, broadcast_shape))
         return reduce_axis
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/oneslike.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/oneslike.py
index e4a4a4d0..7b91cb2a 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/oneslike.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/oneslike.py
@@ -20,7 +20,24 @@ class OnesLike(Expander):
     """OnesLike expander"""
 
     def _expand(self, graph_builder):
+        """
+        将输入张量扩展至指定形状。
+
+        Args:
+            graph_builder: 图构建器对象，用于构建图结构。
+
+        Returns:
+            扩展后的张量。
+
+        """
+        # 获取输入张量
         input_x = self.inputs[0]
+
+        # 创建一个值为1的常量，数据类型与输入张量相同
         const_one = graph_builder.value(input_x.dtype, 1)
+
+        # 使用BroadcastTo操作将常量扩展至输入张量的形状
         result = graph_builder.emit('BroadcastTo', [const_one], attrs={'shape': input_x.shape})
+
+        # 返回扩展后的张量
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/reduce_mean.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/reduce_mean.py
index abb2df57..2258135a 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/reduce_mean.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/reduce_mean.py
@@ -22,22 +22,41 @@ from ._utils import Expander, ExpanderInfoValidator as VLD
 class ReduceMean(Expander):
     """ReduceMean expander"""
 
-    def _expand(self, graph_builder):
+def _expand(self, graph_builder):
+        """
+        对输入张量进行扩展操作。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            Tensor: 扩展操作后的张量。
+
+        """
+        # 获取输入张量
         x = self.inputs[0]
+        # 获取扩展操作的轴
         axis = self.attrs['axis']
+        # 获取是否保持维度
         keep_dims = self.attrs['keep_dims']
 
+        # 如果轴不是元组或列表，则将其转换为元组
         if not isinstance(axis, (tuple, list)):
             axis = (axis,)
+        # 如果轴为空，则将其设置为张量的所有维度
         elif not axis:
             axis = list(range(len(x.shape)))
+        # 计算缩减的大小
         reduce_size = 1.0
         for idx in axis:
             reduce_size *= x.shape[idx]
 
+        # 创建一个与输入张量相同数据类型的值，值为缩减的大小
         reduce_size_value = graph_builder.value(x.dtype, reduce_size)
 
+        # 沿指定轴对输入张量进行求和操作
         sum_x = graph_builder.emit('ReduceSum', [x], attrs={'reduce_axis': axis, 'keep_dims': keep_dims})
+        # 将求和结果除以缩减的大小，得到扩展后的张量
         result = graph_builder.emit('RealDiv', [sum_x, reduce_size_value])
 
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/relu_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/relu_grad.py
index d2e7f740..1fe53f22 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/relu_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/relu_grad.py
@@ -21,12 +21,28 @@ class ReluGrad(Expander):
     """ReLU expander"""
 
     def _expand(self, graph_builder):
+        """
+        在指定的图构建器中扩展当前节点。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器实例，用于在图中生成新的节点。
+
+        Returns:
+            Tensor: 返回计算后的结果张量。
+
+        """
+        # 获取输入张量
         input_x = self.inputs[0]
         input_y = self.inputs[1]
 
+        # 生成一个与input_y相同数据类型的0值张量
+        # 生成一个与input_y相同数据类型的0值张量
         const_zero = graph_builder.value(input_y.dtype, 0)
+        # 判断input_y是否大于0，生成布尔张量
         ge_result = graph_builder.emit('Greater', [input_y, const_zero])
+        # 将布尔张量转换为与input_x相同数据类型的张量
         ge_result = graph_builder.emit('Cast', [ge_result], attrs={'dst_type': input_x.dtype})
+        # 将转换后的张量与input_x相乘
         result = graph_builder.emit('Mul', [ge_result, input_x])
 
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_cross_entropy_with_logits.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_cross_entropy_with_logits.py
index c2c9d20e..c8ed536d 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_cross_entropy_with_logits.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_cross_entropy_with_logits.py
@@ -21,21 +21,46 @@ class SigmoidCrossEntropyWithLogits(Expander):
     """SigmoidCrossEntropyWithLogits expander"""
 
     def _expand(self, graph_builder):
+        """
+        计算sigmoid交叉熵损失。
+
+        Args:
+            graph_builder: 图构建器对象，用于构建计算图。
+
+        Returns:
+            计算得到的sigmoid交叉熵损失值。
+
+        """
         logits, labels = self.inputs
+        # 计算 logits 和 labels 的 sigmoid_cross_entropy_with_logits
         # Calculate sigmoid_cross_entropy_with_logits(logits, labels)
-        # formula of sigmoid_cross_entropy_with_logits is:
-        #     -(labels * log(sigmoid(logits)) + (1 - labels) * log(1 - sigmoid(logits)))
-        # To ensure stability and avoid overflow, the formula equal to :
-        #     max(logits, 0) - logits * labels + log(1 + exp(-abs(logits)))
+        # sigmoid_cross_entropy_with_logits 的公式为：
+        # -(labels * log(sigmoid(logits)) + (1 - labels) * log(1 - sigmoid(logits)))
+        # 为了确保稳定性并避免溢出，该公式等价于：
+        # max(logits, 0) - logits * labels + log(1 + exp(-abs(logits)))
+
+        # 创建一个值为 1.0 的常量
         const_one = graph_builder.value(logits.dtype, 1.0)
+        # 创建一个值为 0.0 的常量
         const_zero = graph_builder.value(logits.dtype, 0.0)
+
+        # 计算 logits 和 0 的最大值
         max_logits = graph_builder.emit('Maximum', [logits, const_zero])
+        # 计算 logits 和 labels 的乘积
         logits_mul_labels = graph_builder.emit('Mul', [logits, labels])
+        # 计算 logits 的绝对值
         abs_logits = graph_builder.emit('Abs', [logits])
+        # 计算 logits 的负值
         neg_abs_logits = graph_builder.emit('Neg', [abs_logits])
+        # 计算 exp(-abs(logits))
         exp_neg_abs_logits = graph_builder.emit('Exp', [neg_abs_logits])
+        # 计算 1 + exp(-abs(logits))
         one_add_exp_neg_abs_logits = graph_builder.emit('Add', [const_one, exp_neg_abs_logits])
+        # 计算 log(1 + exp(-abs(logits)))
         log_one_add_exp_neg_abs_logits = graph_builder.emit('Log', [one_add_exp_neg_abs_logits])
+        # 计算 max(logits, 0) - logits * labels
         res_tmp = graph_builder.emit('Sub', [max_logits, logits_mul_labels])
+        # 计算最终结果
         res = graph_builder.emit('Add', [res_tmp, log_one_add_exp_neg_abs_logits])
+
         return res
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_cross_entropy_with_logits_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_cross_entropy_with_logits_grad.py
index 07f9dba3..cf49d4cc 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_cross_entropy_with_logits_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_cross_entropy_with_logits_grad.py
@@ -21,15 +21,29 @@ class SigmoidCrossEntropyWithLogitsGrad(Expander):
     """SigmoidCrossEntropyWithLogitsGrad expander"""
 
     def _expand(self, graph_builder):
+        """
+        计算sigmoid交叉熵损失的梯度。
+
+        Args:
+            graph_builder: 图构建器对象，用于构建计算图。
+
+        Returns:
+            计算得到的梯度值。
+
+        """
         logits, label, dout = self.inputs
-        # Calculate sigmoid_cross_entropy_with_logits_grad(logits, label, dout)
-        # formula of sigmoid_cross_entropy_with_logits_grad is :
+        # 计算sigmoid_cross_entropy_with_logits_grad(logits, label, dout)
+        # sigmoid_cross_entropy_with_logits_grad的公式为:
         #     (sigmoid(logits) - label) * dout
+        # 计算sigmoid(logits)
+        # Calculate sigmoid(logits)
         const_one = graph_builder.value(logits.dtype, 1.0)
-        neg_x = graph_builder.emit('Neg', [logits])
-        exp_neg_x = graph_builder.emit('Exp', [neg_x])
-        add_exp = graph_builder.emit('Add', [const_one, exp_neg_x])
-        sigmoid_res = graph_builder.emit('RealDiv', [const_one, add_exp])
+        neg_x = graph_builder.emit('Neg', [logits])  # 计算-logits
+        exp_neg_x = graph_builder.emit('Exp', [neg_x])  # 计算e^(-logits)
+        add_exp = graph_builder.emit('Add', [const_one, exp_neg_x])  # 计算1 + e^(-logits)
+        sigmoid_res = graph_builder.emit('RealDiv', [const_one, add_exp])  # 计算1 / (1 + e^(-logits))，即sigmoid(logits)
+        # 计算(sigmoid(logits) - label)
         sigmoid_res_sub_label = graph_builder.emit('Sub', [sigmoid_res, label])
-        res = graph_builder.emit('Mul', [sigmoid_res_sub_label, dout])
+        # 计算最终结果
+        res = graph_builder.emit('Mul', [sigmoid_res_sub_label, dout])  # 计算(sigmoid(logits) - label) * dout
         return res
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_grad.py
index 56d9413a..72564754 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sigmoid_grad.py
@@ -16,16 +16,31 @@
 from ._utils import Expander, ExpanderInfoValidator as VLD
 
 
-@VLD.check_all_formats_same
+@VLD.check_all_formats_same #  定义一个SigmoidGrad类，继承自Expander类
 class SigmoidGrad(Expander):
     """SigmoidGrad expander"""
 
     def _expand(self, graph_builder):
+        """
+        计算 sigmoid 函数的梯度。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于在图中添加操作。
+
+        Returns:
+            Tensor: 计算得到的 sigmoid 梯度。
+
+        """
         input_y, dy = self.inputs
+        # 计算 sigmoid_grad(y, dy)
+        # sigmoid_grad 的公式是: (1 - y) * y * dy
         # Calculate sigmoid_grad(y, dy)
         # formula of sigmoid_grad is : (1 - y) * y * dy
         const_one = graph_builder.value(input_y.dtype, 1.0)
+        # 1 - y
         one_mins_y = graph_builder.emit('Sub', [const_one, input_y])
+        # y * dy
         y_mul_dy = graph_builder.emit('Mul', [input_y, dy])
+        # (1 - y) * (y * dy)
         res = graph_builder.emit('Mul', [one_mins_y, y_mul_dy])
         return res
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/slice.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/slice.py
index 5bc265ea..fb847dd2 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/slice.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/slice.py
@@ -21,15 +21,41 @@ class Slice(Expander):
     """Slice expander"""
 
     def _expand(self, graph_builder):
+        """
+        在图中扩展输入张量。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            Tensor: 扩展后的输出张量。
+
+        """
+        # 获取输入张量
         input_x = self.inputs[0]
+
+        # 获取开始索引
         begin = self.attrs['begin']
+        # 获取切片大小
         size = self.attrs['size']
+
+        # 初始化结束索引列表
         end = []
+        # 初始化步长列表
         strides = []
+
+        # 遍历每个维度，计算结束索引和步长
         for i, begin_idx in enumerate(begin):
+            # 步长设置为1
             strides.append(1)
+            # 计算结束索引
             end.append(begin_idx + size[i])
+
+        # 创建一个新的张量作为输出
         output = graph_builder.tensor(size, input_x.dtype, input_x.data_format)
+
+        # 执行StridedSlice操作，对输入张量进行切片
         graph_builder.op('StridedSlice', output, [input_x], attrs={'begin': begin, 'end': end, 'strides': strides})
 
+        # 返回输出张量
         return output
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax.py
index 991d5e6a..9bad459e 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax.py
@@ -25,45 +25,75 @@ class Softmax(Expander):
     """Softmax expander"""
 
     def _expand(self, graph_builder):
+        """
+        计算Softmax函数值。
+
+        Args:
+            graph_builder: 图构建器对象。
+
+        Returns:
+            Softmax函数的计算结果。
+
+        """
+        # 获取输入数据
         input_x = self.inputs[0]
+        # 获取处理器
         processor = self.processor
+        # 获取轴信息
         axis = self.attrs['axis']
 
+        # 获取输入数据的原始形状
         ori_shape = input_x.shape
+        # 如果输入数据格式为FRAC_NZ，则推断其形状
         if input_x.data_format == DF.FRAC_NZ:
             ori_shape = infer_shape_from_fractalnz(input_x.shape)
 
+        # 遍历轴信息，处理负数轴索引
         for i, _ in enumerate(list(axis)):
             if axis[i] < 0:
                 axis[i] += len(ori_shape)
 
+        # 获取减少维度后的原始形状
         ori_reduced_shape = get_reduced_ori_shape(ori_shape, axis)
 
+        # 获取减少的轴
         reduce_axis = axis
+        # 如果输入数据格式为FRAC_NZ，则转换轴
         if input_x.data_format == DF.FRAC_NZ:
             reduce_axis = to_frac_z_axis(ori_shape, axis)
 
+        # 获取输入数据的原始数据类型
         ori_dtype = input_x.dtype
+        # 如果原始数据类型不是float16且处理器为aicore，则进行类型转换
         if ori_dtype != "float16" and processor == "aicore":
             input_x_f16 = graph_builder.emit('Cast', [input_x], attrs={'dst_type': 'float16'})
             max_x_f16 = graph_builder.emit('ReduceMax', [input_x_f16], attrs={'reduce_axis': reduce_axis,
                                                                               'keep_dims': True})
             max_x = graph_builder.emit('Cast', [max_x_f16], attrs={'dst_type': ori_dtype})
         else:
+            # 计算最大值
             max_x = graph_builder.emit('ReduceMax', [input_x], attrs={'reduce_axis': reduce_axis, 'keep_dims': True})
 
+        # 如果原始数据类型为float16且处理器为aicore，则进行类型转换
         if ori_dtype == "float16" and processor == "aicore":
             max_x = graph_builder.emit('Cast', [max_x], attrs={'dst_type': "float32"})
             input_x = graph_builder.emit('Cast', [input_x], attrs={'dst_type': "float32"})
 
+        # 如果输入数据格式为FRAC_NZ，则重新调整最大值的形状
         if input_x.data_format == DF.FRAC_NZ:
             max_x = graph_builder.emit('Reshape', [max_x], attrs={'shape': ori_reduced_shape})
+        # 计算输入数据减去最大值的差值
         data_sub = graph_builder.emit('Sub', [input_x, max_x])
+        # 计算差值的指数
         data_exp = graph_builder.emit('Exp', [data_sub])
+        # 计算指数的和
         data_expsum = graph_builder.emit('ReduceSum', [data_exp], attrs={'reduce_axis': reduce_axis, 'keep_dims': True})
+        # 如果输入数据格式为FRAC_NZ，则重新调整指数和的形状
         if input_x.data_format == DF.FRAC_NZ:
             data_expsum = graph_builder.emit('Reshape', [data_expsum], attrs={'shape': ori_reduced_shape})
+        # 计算Softmax值
         result = graph_builder.emit('RealDiv', [data_exp, data_expsum])
+        # 如果原始数据类型为float16且处理器为aicore，则进行类型转换
         if ori_dtype == "float16" and processor == "aicore":
             result = graph_builder.emit('Cast', [result], attrs={'dst_type': ori_dtype})
 
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax_cross_entropy_with_logits.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax_cross_entropy_with_logits.py
index e28e74f4..12cae367 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax_cross_entropy_with_logits.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax_cross_entropy_with_logits.py
@@ -22,21 +22,45 @@ class SoftmaxCrossEntropyWithLogits(Expander):
     """SoftmaxCrossEntropyWithLogits expander"""
 
     def _expand(self, graph_builder):
+        """
+        计算损失值和 logits 的梯度。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            Tuple[Tensor, Tensor]: 损失值和 logits 的梯度。
+
+        """
         logits, label = self.inputs
+        # 计算 softmax_cross_entropy_with_logits(logits, label)
+        # softmax_cross_entropy_with_logits 的公式是： -reduce_sum(label * log(softmax(logits)))
         # Calculate softmax_cross_entropy_with_logits(logits, label)
         # formula of softmax_cross_entropy_with_logits is : -reduce_sum(label * log(softmax(logits)))
         axis = (-1,)
         max_x = graph_builder.emit('ReduceMax', [logits], attrs={'reduce_axis': axis, 'keep_dims': True})
+        # 计算 logits 的最大值
         data_sub = graph_builder.emit('Sub', [logits, max_x])
+        # logits 减去最大值
         data_exp = graph_builder.emit('Exp', [data_sub])
+        # 对上一步结果进行指数运算
         data_expsum = graph_builder.emit('ReduceSum', [data_exp], attrs={'reduce_axis': axis, 'keep_dims': True})
+        # 对指数运算结果求和
         data_softmax = graph_builder.emit('RealDiv', [data_exp, data_expsum])
+        # 计算 softmax
         const_eps = graph_builder.value(logits.dtype, 0.000001)
+        # 定义一个极小的常数，用于防止除以零的错误
         data_softmax_safety = graph_builder.emit("Maximum", [data_softmax, const_eps])
+        # 确保 softmax 的值不为零
         softmax_log = graph_builder.emit('Log', [data_softmax_safety])
+        # 对 softmax 结果取对数
         label_mul_log = graph_builder.emit('Mul', [label, softmax_log])
+        # 将 label 与 softmax 的对数相乘
         tmp_res = data_expsum = graph_builder.emit('ReduceSum', [label_mul_log], attrs={
             'reduce_axis': axis, 'keep_dims': False})
+        # 对上一步结果进行求和
         loss = graph_builder.emit('Neg', [tmp_res])
+        # 计算损失值，即上一步结果的负值
         dlogits = graph_builder.emit('Sub', [data_softmax, label])
+        # 计算 logits 的梯度
         return loss, dlogits
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax_grad_ext.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax_grad_ext.py
index d4ccff60..d3c23a4e 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax_grad_ext.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/softmax_grad_ext.py
@@ -13,29 +13,48 @@
 # limitations under the License.
 # ===========================================================================
 """generate json desc for SoftmaxGradExt"""
-from mindspore._extends.graph_kernel.model.model import DataFormat as DF
-from ._utils import Expander, ExpanderInfoValidator as VLD
-from ._utils import get_reduce_axis_shape
+from mindspore._extends.graph_kernel.model.model import DataFormat as DF #  导入DataFormat类
+from ._utils import Expander, ExpanderInfoValidator as VLD #  导入Expander和ExpanderInfoValidator类
+from ._utils import get_reduce_axis_shape #  导入get_reduce_axis_shape函数
 
 
-@VLD.add_format(DF.FRAC_NZ, DF.FRAC_NZ, DF.DEFAULT)
+@VLD.add_format(DF.FRAC_NZ, DF.FRAC_NZ, DF.DEFAULT) #  使用ExpanderInfoValidator类添加FRAC_NZ格式
 @VLD.add_format(DF.DEFAULT, DF.DEFAULT, DF.DEFAULT)
 @VLD.check_attrs('axis')
 class SoftmaxGradExt(Expander):
     """SoftmaxGradExt expander"""
 
     def _expand(self, graph_builder):
+        """
+        对输入数据进行扩展处理。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            Tensor: 处理后的数据。
+
+        """
+        # 获取输入参数
         x, y, z = self.inputs
+        # 获取指定的轴
         axis = self.attrs['axis']
 
+        # 获取需要减少的轴和原始减少的形状
         reduce_axis, ori_reduced_shape = get_reduce_axis_shape(x.shape, x.data_format, axis)
 
+        # 将x和y相乘
         data_mul = graph_builder.emit('Mul', [x, y])
+        # 对乘积进行求和，并保留维度
         data_sum = graph_builder.emit('ReduceSum', [data_mul],
                                       attrs={'reduce_axis': reduce_axis, 'keep_dims': True, 'reduce_output_fuse': True})
+        # 如果x的数据格式为FRAC_NZ，则对求和结果进行重塑
         if x.data_format == DF.FRAC_NZ:
             data_sum = graph_builder.emit('Reshape', [data_sum], attrs={'shape': ori_reduced_shape})
+        # 从x中减去求和结果
         data_sub = graph_builder.emit('Sub', [x, data_sum])
+        # 将减法结果与y相乘
         data_mul2 = graph_builder.emit('Mul', [data_sub, y])
+        # 将结果与z相乘得到最终结果
         result = graph_builder.emit('Mul', [data_mul2, z])
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sqrt_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sqrt_grad.py
index 9f072e7a..8d1bb0bd 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sqrt_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/sqrt_grad.py
@@ -21,9 +21,24 @@ class SqrtGrad(Expander):
     """SqrtGrad expander"""
 
     def _expand(self, graph_builder):
+        """
+        计算并返回给定输入 x 的平方根的梯度。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于构建计算图。
+
+        Returns:
+            Tensor: 返回给定输入 x 的平方根的梯度。
+
+        """
+        # 获取输入 x 和梯度 dout
         # formula of sqrt_grad is dout / (2 * x)
         x, dout = self.inputs
+        # 创建一个常数值 2
         const_two = graph_builder.value(x.dtype, 2)
+        # 计算 2 * x
         dividend = graph_builder.emit('Mul', [x, const_two])
+        # 计算梯度：dout / (2 * x)
         result = graph_builder.emit('RealDiv', [dout, dividend])
+        # 返回计算结果
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/square_sum_all.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/square_sum_all.py
index a06c1a39..f38953d8 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/square_sum_all.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/square_sum_all.py
@@ -21,24 +21,57 @@ class SquareSumAll(Expander):
     """SquareSumAll expander"""
 
     def _check(self):
+        """
+        检查输入是否合法。
+
+        Args:
+            无。
+
+        Returns:
+            无。
+
+        Raises:
+            GKException: 如果输入的数量不等于2，则抛出GKException异常。
+
+        """
         """check inputs"""
+        # 获取输入的数量
         input_num = len(self.inputs)
         if input_num != 2:
+            # 如果输入的数量不等于2，则抛出异常
             raise GKException("For 'SquareSumAll', the inputs number should be 2, but got {}.".format(input_num))
 
     def _expand(self, graph_builder):
+        """
+        对输入的两个变量进行扩展操作。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于在图构建过程中发射操作。
+
+        Returns:
+            tuple: 包含两个元素的元组，每个元素为扩展操作的结果。
+
+        """
         """do expand"""
+        # 获取输入的两个变量
         x0 = self.inputs[0]
         x1 = self.inputs[1]
 
+        # 获取x0的形状
         ori_shape = x0.shape
+        # 初始化一个空列表，用于存储维度索引
         axis = []
+        # 遍历ori_shape，将每个维度的索引添加到axis列表中
         for i, _ in enumerate(ori_shape):
             axis.append(i)
 
+        # 对x0进行平方运算
         square_res0 = graph_builder.emit('Mul', [x0, x0])
+        # 对x1进行平方运算
         square_res1 = graph_builder.emit('Mul', [x1, x1])
+        # 对square_res0进行求和运算，求和的维度为axis，并保持维度不变
         result0 = graph_builder.emit('ReduceSum', [square_res0], attrs={'reduce_axis': axis, 'keep_dims': False})
+        # 对square_res1进行求和运算，求和的维度为axis，并保持维度不变
         result1 = graph_builder.emit('ReduceSum', [square_res1], attrs={'reduce_axis': axis, 'keep_dims': False})
 
         return result0, result1
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/square_sum_v1.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/square_sum_v1.py
index 8774dcbe..67d6f28b 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/square_sum_v1.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/square_sum_v1.py
@@ -25,13 +25,30 @@ class SquareSumV1(Expander):
     """Square expander"""
 
     def _expand(self, graph_builder):
+        """
+        计算输入张量的平方并沿指定轴进行求和。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象。
+
+        Returns:
+            Tensor: 计算得到的张量。
+
+        """
+        # 获取输入的第一个元素
         x = self.inputs[0]
+        # 获取属性中的axis值
         axis = self.attrs['axis']
 
+        # 获取需要reduce的axis和原始的reduced shape
         reduce_axis, ori_reduced_shape = get_reduce_axis_shape(x.shape, x.data_format, axis)
 
+        # 计算x的平方
         square_res = graph_builder.emit('Mul', [x, x])
+        # 对平方结果进行ReduceSum操作
         result = graph_builder.emit('ReduceSum', [square_res], attrs={'reduce_axis': reduce_axis, 'keep_dims': False})
+        # 如果数据格式为FRAC_NZ，则对结果进行Reshape操作
         if x.data_format == DF.FRAC_NZ:
             result = graph_builder.emit('Reshape', [result], attrs={'shape': ori_reduced_shape})
+        # 返回最终结果
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/squared_difference.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/squared_difference.py
index 316b000e..ca0dc687 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/squared_difference.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/squared_difference.py
@@ -21,10 +21,24 @@ class SquaredDifference(Expander):
     """SquaredDifference expander"""
 
     def _expand(self, graph_builder):
+        """
+        根据输入的两个输入值计算并返回它们的平方差的计算结果。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于在图中生成节点和边。
+
+        Returns:
+            Node: 计算结果节点。
+
+        """
+        # 获取输入的第一个值
         input_x = self.inputs[0]
+        # 获取输入的第二个值
         input_y = self.inputs[1]
 
+        # 使用图构建器计算输入值的差值
         sub_val = graph_builder.emit('Sub', [input_x, input_y])
+        # 使用图构建器计算差值的平方
         result = graph_builder.emit('Mul', [sub_val, sub_val])
 
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/squeeze.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/squeeze.py
index a5e631b5..66178c0b 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/squeeze.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/squeeze.py
@@ -21,27 +21,67 @@ class Squeeze(Expander):
     """Squeeze expander"""
 
     def _expand(self, graph_builder):
+        """
+        扩展输入的维度。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于构建图结构。
+
+        Returns:
+            Tensor: 扩展维度后的输入。
+
+        """
+        # 获取输入的第一个元素
         input_x = self.inputs[0]
+
+        # 根据输入的shape和axis属性推断输出shape
         out_shape = self.infer_shape(input_x.shape, self.attrs['axis'])
+
+        # 使用graph_builder发射Reshape操作，并设置shape属性为out_shape
         result = graph_builder.emit('Reshape', [input_x], attrs={'shape': out_shape})
 
+        # 返回结果
         return result
 
     @staticmethod
     def infer_shape(shape, axis):
+        """
+        根据指定的axis推断squeeze后的shape。
+
+        Args:
+            shape (list, tuple): 原始数据的shape。
+            axis (int, list, tuple): 需要被squeeze的维度。如果为int，则只squeeze该维度；
+                如果为list或tuple，则squeeze列表或元组中的每个维度。如果为空，则squeeze所有维度为1的维度。
+
+        Returns:
+            list: squeeze后的shape。
+
+        Raises:
+            ValueError: 如果输入的axis类型不符合要求，抛出异常。
+
+        """
         """infer shape for squeeze"""
         def squeeze_axis(shape, axis):
+            # 如果axis为空，移除shape中所有值为1的维度
             if not axis:
                 out_shape = list(d for d in shape if d != 1)
             else:
+                # 获取shape的维度数量
                 ndim = len(shape)
+                # 移除shape中指定的axis维度
                 out_shape = list(shape[i] for i in range(ndim) if not (i in axis or (i - ndim) in axis))
+            # 如果out_shape为空，则将其设置为[1]
             if not out_shape:
                 out_shape = [1]
             return out_shape
+
+        # 如果shape是列表或元组类型
         if isinstance(shape, (list, tuple)):
+            # 如果axis是整数类型，则将其转换为列表
             if isinstance(axis, int):
                 axis = [axis]
+            # 如果axis是列表或元组类型，则调用squeeze_axis函数处理
             if isinstance(axis, (list, tuple)):
                 return squeeze_axis(shape, axis)
+        # 如果输入不符合要求，则抛出异常
         raise ValueError("Invalid axis for Squeeze.")
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/tanh_grad.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/tanh_grad.py
index a43c7962..60681a54 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/tanh_grad.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/tanh_grad.py
@@ -21,11 +21,29 @@ class TanhGrad(Expander):
     """TanhGrad expander"""
 
     def _expand(self, graph_builder):
+        """
+        计算1减去输入值的平方后，再与输入的导数相乘。
+
+        Args:
+            graph_builder (GraphBuilder): 图构建器对象，用于构建计算图。
+
+        Returns:
+            Tensor: 计算结果，类型为Tensor。
+
+        """
+        # 获取输入值
         input_y, input_dy = self.inputs
 
+        # 创建一个值为1的常量，数据类型与input_y相同
         const_one = graph_builder.value(input_y.dtype, 1)
+
+        # 计算input_y的平方
         double_y = graph_builder.emit('Mul', [input_y, input_y])
+
+        # 计算1减去input_y的平方
         one_sub_double_y = graph_builder.emit('Sub', [const_one, double_y])
+
+        # 计算input_dy与1减去input_y的平方的乘积
         result = graph_builder.emit('Mul', [input_dy, one_sub_double_y])
 
         return result
diff --git a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/tile.py b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/tile.py
index 918b7486..9b262d45 100644
--- a/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/tile.py
+++ b/src/mindspore2022/mindspore/python/mindspore/_extends/graph_kernel/expanders/tile.py
@@ -25,30 +25,48 @@ class Tile(Expander):
 
     def _get_output_shape(self):
         """Get output shape"""
+        # 获取输入形状的列表
         shape = list(self.inputs[0].shape)
+        # 获取属性"multiples"的列表
         multiples = list(self.attrs["multiples"])
 
+        # 计算"multiples"和输入形状的长度差
         diff_len = len(multiples) - len(shape)
+        # 如果长度差小于0，抛出异常
         if diff_len < 0:
             raise GKException("For 'Tile', dimensions of attr 'multiples' should be greater than or equal to "
                               "dimensions of input shape, but got {} and {}".format(multiples, shape))
+        # 如果长度差大于0，则扩展输入形状的列表
         if diff_len > 0:
             for _ in range(diff_len):
                 shape.insert(0, 1)
 
+        # 初始化输出形状的列表
         output_shape = []
 
+        # 遍历输入形状和multiples的元组
         for sh, mul in list(zip(shape, multiples)):
+            # 如果输入形状和multiples的值都不为1，则抛出异常
             if sh != 1 and mul != 1:
                 raise GKException("For 'Tile', input shape{} and attr 'multiples'{} can not broadcast."
                                   .format(self.inputs[0].shape, multiples))
+            # 计算维度
             dim = sh * mul
+            # 将计算得到的维度添加到输出形状的列表中
             output_shape.append(dim)
+        # 返回输出形状的列表
         return output_shape
 
     def _expand(self, graph_builder):
+        # 获取输入的第一个元素
         input_x = self.inputs[0]
+
+        # 获取输出形状
         output_shape = self._get_output_shape()
 
+        # 使用graph_builder的emit方法生成BroadcastTo操作
+        # 参数为[input_x]和输出形状
         result = graph_builder.emit('BroadcastTo', [input_x], attrs={'shape': output_shape})
+    
+        # 返回结果
         return result