mindspore 2.1.0__cp38-cp38-win_amd64.whl → 2.2.11__cp38-cp38-win_amd64.whl

mindspore/ops/operations/nn_ops.py

@@ -1,4 +1,4 @@
-# Copyright 2020-2022 Huawei Technologies Co., Ltd
+# Copyright 2020-2023 Huawei Technologies Co., Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -438,7 +438,10 @@ class Softmax(Primitive):
 
     Inputs:
         - **logits** (Tensor) - Tensor of shape :math:`(N, *)`, where :math:`*` means, any number of
-          additional dimensions, with float16, float32 or float64(CPU, GPU) data type.
+          additional dimensions. Supported dtypes:
+
+          - Ascend: float16, float32.
+          - GPU/CPU: float16, float32, float64.
 
     Outputs:
         Tensor, with the same type and shape as the logits.
@@ -517,7 +520,11 @@ class Softplus(Primitive):
         \text{output} = \log(1 + \exp(\text{x}))
 
     Inputs:
-        - **input_x** (Tensor) - Tensor of any dimension, with float16, float32 or float64(CPU, GPU) data type.
+        - **input_x** (Tensor) - Tensor of any dimension.
+          Supported dtypes:
+
+          - GPU/CPU: float16, float32, float64.
+          - Ascend: float16, float32.
 
     Outputs:
         Tensor, with the same type and shape as the `input_x`.
@@ -626,7 +633,7 @@ class ReLUV3(Primitive):
     Inputs:
         - **input_x** (Tensor) - Tensor of shape :math:`(N, *)`, where :math:`*` means, any number of
           additional dimensions, data type is
-          `number <https://www.mindspore.cn/docs/en/r2.1/api_python/mindspore.html#mindspore.dtype>`_.
+          `number <https://www.mindspore.cn/docs/en/r2.2/api_python/mindspore.html#mindspore.dtype>`_.
 
     Outputs:
         Tensor of shape :math:`(N, *)`, with the same type and shape as the `input_x`.
@@ -659,7 +666,11 @@ class Mish(PrimitiveWithInfer):
     Refer to :func:`mindspore.ops.mish` for more details.
 
     Inputs:
-        - **x** (Tensor) - The input Tensor with float16, float32 or float64 data type.
+        - **x** (Tensor) - The input Tensor.
+          Supported dtypes:
+
+          - GPU/CPU: float16, float32, float64.
+          - Ascend: float16, float32.
 
     Outputs:
         Tensor, with the same type and shape as the `x`.
@@ -745,7 +756,9 @@ class ReLU6(PrimitiveWithCheck):
     Refer to :func:`mindspore.ops.relu6` for more details.
 
     Inputs:
-        - **input_x** (Tensor) - Input Tensor of float16 or float32 data type.
+        - **input_x** (Tensor) - Tensor of shape :math:`(N, *)`,
+          where :math:`*` means any number of additional dimensions.
+          Data type must be float16, float32.
 
     Outputs:
         Tensor, with the same type and shape as the `input_x`.
@@ -1216,54 +1229,6 @@ class InstanceNormV2(Primitive):
         validator.check_bool(is_training, "is_training", self.name)
 
 
-class BNTrainingReduce(Primitive):
-    """
-    The BNTrainingReduce interface is deprecated, please use the :class:`mindspore.ops.BatchNorm` instead.
-
-    Supported Platforms:
-        Deprecated
-    """
-
-    @deprecated("1.5", "ops.BatchNorm", False)
-    @prim_attr_register
-    def __init__(self, data_format="NCHW"):
-        """Initialize BNTrainingReduce."""
-        self.init_prim_io_names(inputs=['x'], outputs=['sum', 'square_sum'])
-        self.format = validator.check_string(data_format, ['NCHW', 'NHWC'], 'format', self.name)
-        if context.get_context("device_target") != "GPU" and self.format == "NHWC":
-            raise ValueError(f"For '{self.name}', the 'NHWC' format is only supported in GPU target, "
-                             f"but got the 'data_format' is {self.format} and "
-                             f"the platform is {context.get_context('device_target')}.")
-        self.add_prim_attr('data_format', self.format)
-
-
-class BNTrainingUpdate(Primitive):
-    """
-    The BNTrainingUpdate interface is deprecated, please use the :class:`mindspore.ops.BatchNorm` instead.
-
-    Supported Platforms:
-        Deprecated
-    """
-
-    @deprecated("1.5", "ops.BatchNorm", False)
-    @prim_attr_register
-    def __init__(self, isRef=True, epsilon=1e-5, factor=0.1, data_format="NCHW"):
-        """Initialize BNTrainingUpdate."""
-        self.init_prim_io_names(inputs=['x', 'sum', 'square_sum', 'scale', 'b', 'mean', 'variance'],
-                                outputs=['y', 'running_mean', 'running_variance', 'save_mean', 'save_inv_variance'])
-        validator.check_value_type("isRef", isRef, [bool], self.name)
-        validator.check_value_type("epsilon", epsilon, [float], self.name)
-        validator.check_value_type("factor", factor, [float], self.name)
-        self.epsilon = validator.check_float_range(epsilon, 0, 1, validator.INC_RIGHT, 'epsilon', 'BNTrainingUpdate')
-        self.factor = validator.check_float_range(factor, 0, 1, validator.INC_BOTH, 'factor', 'BNTrainingUpdate')
-        self.format = validator.check_string(data_format, ['NCHW', 'NHWC'], 'format', self.name)
-        if context.get_context("device_target") != "GPU" and self.format == "NHWC":
-            raise ValueError(f"For '{self.name}', the 'NHWC' format is only supported in GPU target, "
-                             f"but got the 'data_format' is {self.format} and "
-                             f"the platform is {context.get_context('device_target')}.")
-        self.add_prim_attr('data_format', self.format)
-
-
 class BatchNorm(PrimitiveWithInfer):
     r"""
     Batch Normalization for input data and updated parameters.
@@ -1400,33 +1365,40 @@ class Conv2D(Primitive):
     2D convolution layer.
 
     Applies a 2D convolution over an input tensor which is typically of shape :math:`(N, C_{in}, H_{in}, W_{in})`,
-    where :math:`N` is batch size, :math:`C` is channel number, :math:`H` is height, :math:`W` is width,
-    :math:`X_i` is
-    the :math:`i^{th}` input value and :math:`b_i` indicates the deviation value of the :math:`i^{th}` input value.
-    For each batch of shape :math:`(C_{in}, H_{in}, W_{in})`, the formula is defined as:
+    where :math:`N` is batch size, :math:`C` is channel number, :math:`H` is feature height, :math:`W` is feature width.
+
+    The output is calculated based on formula:
 
     .. math::
 
-        out_j = \sum_{i=0}^{C_{in} - 1} ccor(W_{ij}, X_i) + b_j,
-
-    where :math:`ccor` is the cross correlation operator, :math:`C_{in}` is the input channel number, :math:`j` ranges
-    from :math:`0` to :math:`C_{out} - 1`, :math:`W_{ij}` corresponds to the :math:`i`-th channel of the :math:`j`-th
-    filter and :math:`out_{j}` corresponds to the :math:`j`-th channel of the output. :math:`W_{ij}` is a slice
-    of kernel and it has shape :math:`(\text{kernel_size[0]}, \text{kernel_size[1]})`,
-    where :math:`\text{kernel_size[0]}` and :math:`\text{kernel_size[1]}` are the height and width of the
-    convolution kernel. The full kernel has shape
-    :math:`(C_{out}, C_{in} / \text{group}, \text{kernel_size[0]}, \text{kernel_size[1]})`,
-    where group is the group number to split the input in the channel dimension.
-
-    If the 'pad_mode' is set to be "pad", the output height and width will be
-    :math:`\left \lfloor{1 + \frac{H_{in} + \text{padding[0]} + \text{padding[1]} - \text{kernel_size[0]} -
-    (\text{kernel_size[0]} - 1) \times (\text{dilation[0]} - 1) }{\text{stride[0]}}} \right \rfloor` and
-    :math:`\left \lfloor{1 + \frac{W_{in} + \text{padding[2]} + \text{padding[3]} - \text{kernel_size[1]} -
-    (\text{kernel_size[1]} - 1) \times (\text{dilation[1]} - 1) }{\text{stride[1]}}} \right \rfloor` respectively.
-    Where :math:`dilation` is Spacing between kernel elements, :math:`stride` is The step length of each step,
-    :math:`padding` is zero-padding added to both sides of the input.
-
-    The first introduction can be found in paper `Gradient Based Learning Applied to Document Recognition
+        \text{out}(N_i, C_{\text{out}_j}) = \text{bias}(C_{\text{out}_j}) +
+        \sum_{k = 0}^{C_{in} - 1} \text{ccor}({\text{weight}(C_{\text{out}_j}, k), \text{X}(N_i, k)})
+
+    where :math:`bias` is the output channel bias, :math:`ccor` is
+    the `cross-correlation <https://en.wikipedia.org/wiki/Cross-correlation>`_,
+    , :math:`weight` is the convolution kernel value and :math:`X` represents the input feature map.
+
+    Here are the indices' meanings:
+    - :math:`i` corresponds to the batch number, ranging from 0 to N-1, where N is the batch size of the input.
+
+    - :math:`j` corresponds to the output channel, ranging from 0 to C_{out}-1, where C_{out} is the number of
+      output channels, which is also equal to the number of kernels.
+
+    - :math:`k` corresponds to the input channel, ranging from 0 to C_{in}-1, where C_{in} is the number of
+      input channels, which is also equal to the number of channels in the convolutional kernels.
+
+    Therefore, in the above formula, :math:`{bias}(C_{out_j})` represents the bias of the :math:`j`-th
+    output channel, :math:`{weight}(C_{out_j}, k)` represents the slice of the :math:`j`-th convolutional
+    kernel in the :math:`k`-th channel, and :math:`{X}(N_i, k)` represents the slice of the :math:`k`-th input
+    channel in the :math:`i`-th batch of the input feature map.
+
+    The shape of the convolutional kernel is given by :math:`(kernel\_size[0], kernel\_size[1])`,
+    where :math:`kernel\_size[0]` and :math:`kernel\_size[1]` are the height and width of the kernel, respectively.
+    If we consider the input and output channels as well as the `group` parameter, the complete kernel shape
+    will be :math:`(C_{out}, C_{in} / \text{group}, \text{kernel_size[0]}, \text{kernel_size[1]})`,
+    where `group` is the number of groups dividing `x`'s input channel when applying group convolution.
+
+    For more details about convolution layer, please refer to `Gradient Based Learning Applied to Document Recognition
     <http://vision.stanford.edu/cs598_spring07/papers/Lecun98.pdf>`_.
 
     Note:
@@ -1434,57 +1406,72 @@ class Conv2D(Primitive):
         That is, when `group>1`, condition `in\_channels` = `out\_channels` = `group` must be satisfied.
 
     Args:
-        out_channel (int): The number of output channel :math:`C_{out}`.
-        kernel_size (Union[int, tuple[int]]): The data type is int or a tuple of 2 integers. Specifies the height
-            and width of the 2D convolution window. Single int means the value is for both the height and the width of
-            the kernel. A tuple of 2 ints means the first value is for the height and the other is for the
-            width of the kernel.
-        mode (int): Modes for different convolutions. The value is currently not used. Default: ``1`` .
-        pad_mode (str): Specifies padding mode. The optional values are
-            ``"same"`` , ``"valid"`` and ``"pad"`` . Default: ``"valid"`` .
-
-            - ``"same"``: Adopts the way of completion. The height and width of the output will be equal to
-              the input `x` divided by stride. The padding will be evenly calculated in top and bottom,
-              left and right possiblily.
-              Otherwise, the last extra padding will be calculated from the bottom and the right side.
+        out_channel (int): Specifies output channel :math:`C_{out}`.
+        kernel_size (Union[int, tuple[int]]): Specifies the height and width of the 2D convolution kernel.
+            It can be a single int or a tuple of 2 integers. A single int means the value is for both the height
+            and the width. A tuple of 2 ints means the first value is for the height and the other is for the width.
+        mode (int, optional): Modes for different convolutions. The value is currently not used. Default: ``1`` .
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` , ``"valid"`` or ``"pad"`` . Default: ``"valid"`` .
+
+            - ``"same"``: Pad the input around its edges so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally, If the amount is even, it is
+              uniformly distributed around the input, if it is odd, the excess amount goes to the right/bottom side.
               If this mode is set, `pad` must be 0.
-
-            - ``"valid"``: Adopts the way of discarding. The possible largest height and width of output will be
-              returned without padding. Extra pixels will be discarded. If this mode is set, `pad` must be 0.
-
-            - ``"pad"``: Implicit paddings on both sides of the input `x`. The number of `pad` will be padded to the
-              input Tensor borders. `pad` must be greater than or equal to 0.
-        pad (Union(int, tuple[int])): Implicit paddings on both sides of the input `x`. If `pad` is one integer,
-                    the paddings of top, bottom, left and right are the same, equal to pad. If `pad` is a tuple
-                    with four integers, the paddings of top, bottom, left and right will be equal to pad[0],
-                    pad[1], pad[2], and pad[3] accordingly. Default: ``0`` .
-        stride (Union(int, tuple[int])): The distance of kernel moving, an int number that represents
-            the height and width of movement are both strides, or a tuple of two or four int numbers that
-            represent height and width of movement respectively. Default: ``1`` .
-        dilation (Union(int, tuple[int])): The data type is int or a tuple of 2 or 4 integers. Specifies the dilation
-                                      rate to use for dilated convolution. If set to be :math:`k > 1`, there will
-                                      be :math:`k - 1` pixels skipped for each sampling location. Its value must
-                                      be greater than or equal to 1 and bounded by the height and width of the
-                                      input `x`. Default: ``1`` .
-        group (int): Splits input into groups. Default: ``1`` .
-        data_format (str): The optional value for data format, is ``'NHWC'`` or ``'NCHW'`` . Default: ``"NCHW"`` .
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible height and width. Extra pixels that could not complete a full stride will
+              be discarded. If this mode is set, `pad` must be 0.
+            - ``"pad"``: Pad the input with a specified amount. In this mode, the amount of padding
+              in the height and width directions is determined by the `pad` parameter.
+              If this mode is set, `pad` must be greater than or equal to 0.
+
+        pad (Union(int, tuple[int]), optional): Specifies the amount of padding to apply on input
+            when `pad_mode` is set to ``"pad"``. It can be a single int or a tuple of 4 ints.
+            If `pad` is one integer, the paddings of top, bottom, left and right are the same, equal to `pad`.
+            If `pad` is a tuple with four integers, the paddings of top, bottom, left and right will be equal to pad[0],
+            pad[1], pad[2], and pad[3] accordingly. Default: ``0`` .
+        stride (Union(int, tuple[int]), optional): Specifies the stride of the convolution kernel's movement.
+            It can be a single int or a tuple of two or four ints. A single int means the stride is the same in
+            both the height and width directions. A tuple of two ints indicates the strides in the height and
+            width directions, respectively. For a tuple of four ints, the two ints correspond to (N, C) dimension
+            are treated as 1, and the two correspond to (H, W) dimensions is the step size in the height
+            and width directions respectively. Default: ``1`` .
+        dilation (Union(int, tuple[int]), optional): Specifies the dilation rate to use for dilated convolution.
+            It can be a single int or a tuple of 2 or 4 integers. A single int means the dilation size is the same
+            in both the height and width directions. A tuple of two ints represents the dilation size in
+            the height and width directions, respectively. For a tuple of four ints, the two ints correspond
+            to (N, C) dimension are treated as 1, and the two correspond to (H, W) dimensions is the
+            dilation size in the height and width directions respectively.
+            Assuming :math:`dilation=(d0, d1)`, the convolutional kernel samples the input with a
+            spacing of :math:`d0-1` elements in the height direction and :math:`d1-1` elements in the width direction.
+            The values in the height and width dimensions are in the ranges [1, H] and [1, W], respectively.
+            Default: ``1`` .
+        group (int, optional): Specifies the number of groups dividing `x`'s input channel when applying
+            group convolution. Default: ``1`` .
+        data_format (str, optional): The optional value for data format, is ``'NHWC'`` or ``'NCHW'`` .
+            Default: ``"NCHW"`` .
 
     Inputs:
-        - **x** (Tensor) - Tensor of shape :math:`(N, C_{in}, H_{in}, W_{in})`.
-        - **weight** (Tensor) - Set size of kernel is :math:`(\text{kernel_size[0]}, \text{kernel_size[1]})`,
-          then the shape is :math:`(C_{out}, C_{in}, \text{kernel_size[0]}, \text{kernel_size[1]})`.
+        - **x** (Tensor) - Input tensor of shape :math:`(N, C_{in}, H_{in}, W_{in})` or
+          :math:`(N, H_{in}, W_{in}, C_{in}, )` depending on `data_format` .
+        - **weight** (Tensor) - The convolutional kernel value, it should has shape
+          :math:`(C_{out}, C_{in} / \text{group}, \text{kernel_size[0]}, \text{kernel_size[1]})` .
 
     Outputs:
-        Tensor, the value that applied 2D convolution. The shape is :math:`(N, C_{out}, H_{out}, W_{out})`.
+        Tensor, the value that applied 2D convolution. The shape is :math:`(N, C_{out}, H_{out}, W_{out})`
+        or :math:`(N, H_{out}, W_{out}, C_{out}, )`.
+        To see how different pad modes affect the output shape, please refer to
+        :class:`mindspore.nn.Conv2d` for more details.
 
     Raises:
         TypeError: If `kernel_size`, `stride`, `pad` or `dilation` is neither an int nor a tuple.
         TypeError: If `out_channel` or `group` is not an int.
         ValueError: If `kernel_size`, `stride` or `dilation` is less than 1.
-        ValueError: If `pad_mode` is not one of 'same', 'valid' or 'pad'.
+        ValueError: If `pad_mode` is not one of ``'same'``, ``'valid'`` or ``'pad'``.
         ValueError: If `pad` is a tuple whose length is not equal to 4.
-        ValueError: If `pad_mode` it not equal to 'pad' and `pad` is not equal to (0, 0, 0, 0).
-        ValueError: If `data_format` is neither 'NCHW' nor 'NHWC'.
+        ValueError: If `pad_mode` it not equal to ``'pad'`` and `pad` is not equal to ``(0, 0, 0, 0)``.
+        ValueError: If `data_format` is neither ``'NHWC'`` nor ``'NCHW'`` .
 
     Supported Platforms:
         ``Ascend`` ``GPU`` ``CPU``
@@ -1493,12 +1480,49 @@ class Conv2D(Primitive):
         >>> import mindspore
         >>> import numpy as np
         >>> from mindspore import Tensor, ops
+        >>> # case 1: All parameters use default values.
         >>> x = Tensor(np.ones([10, 32, 32, 32]), mindspore.float32)
         >>> weight = Tensor(np.ones([32, 32, 3, 3]), mindspore.float32)
         >>> conv2d = ops.Conv2D(out_channel=32, kernel_size=3)
         >>> output = conv2d(x, weight)
         >>> print(output.shape)
         (10, 32, 30, 30)
+        >>> # case 2: pad_mode="pad", other parameters being default.
+        >>> x = Tensor(np.ones([10, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([32, 32, 3, 3]), mindspore.float32)
+        >>> conv2d = ops.Conv2D(out_channel=32, kernel_size=3, pad_mode="pad", pad=(4, 10, 4, 10))
+        >>> output = conv2d(x, weight)
+        >>> print(output.shape)
+        (10, 32, 44, 44)
+        >>> # case 3: stride=(2, 4), other parameters being default.
+        >>> x = Tensor(np.ones([10, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([32, 32, 3, 3]), mindspore.float32)
+        >>> conv2d = ops.Conv2D(out_channel=32, kernel_size=3, stride=(2, 4))
+        >>> output = conv2d(x, weight)
+        >>> print(output.shape)
+        (10, 32, 15, 8)
+        >>> # case 4: dilation=2, other parameters being default.
+        >>> x = Tensor(np.ones([10, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([32, 32, 3, 3]), mindspore.float32)
+        >>> conv2d = ops.Conv2D(out_channel=32, kernel_size=3, dilation=2)
+        >>> output = conv2d(x, weight)
+        >>> print(output.shape)
+        (10, 32, 28, 28)
+        >>> # case 5: group=2, other parameters being default.
+        >>> x = Tensor(np.ones([10, 64, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([32, 32, 3, 3]), mindspore.float32)
+        >>> conv2d = ops.Conv2D(out_channel=32, kernel_size=3, group=2)
+        >>> output = conv2d(x, weight)
+        >>> print(output.shape)
+        (10, 32, 30, 30)
+        >>> # case 6: All parameters are specified.
+        >>> x = Tensor(np.ones([10, 64, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([32, 32, 3, 3]), mindspore.float32)
+        >>> conv2d = ops.Conv2D(out_channel=32, kernel_size=3, pad_mode="pad",
+        ...                     pad=(4, 10, 4, 10), stride=(2, 4), dilation=2,  group=2)
+        >>> output = conv2d(x, weight)
+        >>> print(output.shape)
+        (10, 32, 21, 11)
     """
 
     @prim_attr_register
@@ -1779,8 +1803,13 @@ class _Pool(PrimitiveWithInfer):
                 out_w = math.ceil(input_w / stride_w)
         out_shape = [batch, channel, out_h, out_w] if self.format == "NCHW" else [batch, out_h, out_w, channel]
 
-        for shape_value in out_shape:
-            if shape_value <= 0 and shape_value != -1:
+        is_dynamic_shape = False
+        for in_shape_val in x_shape_norm:
+            if in_shape_val == -1:
+                is_dynamic_shape = True
+
+        for out_shape_val in out_shape:
+            if out_shape_val <= 0 and not is_dynamic_shape:
                 raise ValueError(f"For '{self.name}', the each element of the output shape must be larger than 0, "
                                  f"but got output shape: {out_shape}. The input shape: {x_shape}, "
                                  f"kernel size: {self.kernel_size}, strides: {self.strides}."
@@ -1814,22 +1843,26 @@ class MaxPool(_Pool):
         strides (Union[int, tuple[int]]): The distance of kernel moving, an int number that represents
             not only the height of movement but also the width of movement, or a tuple of two int numbers that
             represent height and width of movement respectively. Default: ``1`` .
-        pad_mode (str): The optional value of pad mode is ``"same"`` or ``"valid"`` .
-            Default: ``"valid"`` .
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` or ``"valid"`` . Default: ``"valid"`` .
 
-            - ``"same"``: Adopts the way of completion. The height and width of the output will be the same
-              as the input. The total number of padding will be calculated in horizontal and vertical
-              directions and evenly distributed to top, bottom, left and right if possible.
-              Otherwise, the last extra padding will be done from the bottom and the right side.
+            - ``"same"``: Pad the input around its edges so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally, If the amount is even, it is
+              uniformly distributed around the input, if it is odd, the excess amount goes to the right/bottom side.
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible height and width. Extra pixels that could not complete a full stride will
+              be discarded.
 
-            - ``"valid"``: Adopts the way of discarding. The possible largest height and width of output
-              will be returned without padding. Extra pixels will be discarded.
         data_format (str) : The optional value for data format, is ``'NHWC'`` or ``'NCHW'`` .
             Default: ``'NCHW'`` .
 
     Inputs:
         - **x** (Tensor) - Tensor of shape :math:`(N, C_{in}, H_{in}, W_{in})`.
-          Supported dtypes: float16, float32, float64.
+          Supported dtypes:
+
+          - CPU: float16, float32, float64.
+          - GPU/Ascend: float16, float32.
 
     Outputs:
         Tensor, with shape :math:`(N, C_{out}, H_{out}, W_{out})`.
@@ -1887,16 +1920,17 @@ class MaxPoolV1(Primitive):
         strides (Union[int, tuple[int]]): The distance of kernel moving, an integer that represents
             the height and width of movement are both strides, or a tuple of two integers that
             represent height and width of movement, respectively. Default: ``1`` .
-        pad_mode (str): The optional value for pad mode, is ``"same"`` or ``"valid"`` .
-            Default: ``"valid"`` .
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` or ``"valid"`` . Default: ``"valid"`` .
 
-            - ``"same"``: Adopts the way of completion. The height and width of the output will be the same
-              as the input. The number of padding will be calculated in horizontal and vertical
-              directions, and evenly distributed to top and bottom, left and right if possible.
-              Otherwise, the extra padding will be done from the bottom and the right side.
+            - ``"same"``: Pad the input around its edges so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally, If the amount is even, it is
+              uniformly distributed around the input, if it is odd, the excess amount goes to the right/bottom side.
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible height and width. Extra pixels that could not complete a full stride will
+              be discarded.
 
-            - ``"valid"``: Adopts the way of discarding. The possible largest height and width of the
-              output will be returned without padding. Extra pixels will be discarded.
         data_format (str) : The optional value for data format, is ``'NCHW'`` or ``'NHWC'`` .
             Default: ``'NCHW'`` .
 
@@ -1957,55 +1991,6 @@ class MaxPoolV1(Primitive):
         self.add_prim_attr("strides", strides_adapted)
 
 
-class MaxPoolWithArgmax(Primitive):
-    r"""
-    :class:`mindspore.ops.MaxPoolWithArgmax` is deprecated from version 2.0 and will be removed in a future version,
-    use :class:`mindspore.ops.MaxPoolWithArgmaxV2` instead.
-
-    Supported Platforms:
-        Deprecated
-
-    Examples:
-        >>> import mindspore
-        >>> import numpy as np
-        >>> from mindspore import Tensor, ops
-        >>> x = Tensor(np.arange(1 * 3 * 3 * 4).reshape((1, 3, 3, 4)), mindspore.float32)
-        >>> maxpool_arg_op = ops.MaxPoolWithArgmax(pad_mode="VALID", kernel_size=2, strides=1)
-        >>> output_tensor, argmax = maxpool_arg_op(x)
-        >>> print(output_tensor)
-        [[[[ 5.  6.  7.]
-           [ 9. 10. 11.]]
-          [[17. 18. 19.]
-           [21. 22. 23.]]
-          [[29. 30. 31.]
-           [33. 34. 35.]]]]
-    """
-
-    @deprecated("2.0", "ops.MaxPoolWithArgmaxV2", False)
-    @prim_attr_register
-    def __init__(self, kernel_size=1, strides=1, pad_mode="valid", data_format="NCHW"):
-        """Initialize MaxPoolWithArgmax."""
-        self.init_prim_io_names(inputs=['x'], outputs=['output', 'mask'])
-        validator.check_value_type('kernel_size', kernel_size, [int, tuple], self.name)
-        validator.check_value_type('strides', strides, [int, tuple], self.name)
-        validator.check_value_type('pad_mode', pad_mode, [str], self.name)
-        self.pad_mode = validator.check_string(pad_mode.upper(), ['VALID', 'SAME'], 'pad_mode', self.name)
-        self.add_prim_attr("pad_mode", self.pad_mode)
-        self.format = validator.check_string(data_format, ['NCHW', 'NHWC'], 'format', self.name)
-        if context.get_context("device_target") != "GPU" and self.format == "NHWC":
-            raise ValueError(f"For '{self.name}', the 'NHWC' format is only supported in GPU target, "
-                             f"but got the 'data_format' is {self.format} and "
-                             f"the platform is {context.get_context('device_target')}.")
-        self.kernel_size = _check_positive_int_or_tuple(
-            "kernel_size", kernel_size, self.name, allow_four=False, ret_four=True)
-        self.kernel_size = (1, self.kernel_size[-2], self.kernel_size[-1], 1)
-        self.add_prim_attr("kernel_size", self.kernel_size)
-
-        self.strides = _check_positive_int_or_tuple("strides", strides, self.name, allow_four=False, ret_four=True)
-        self.strides = (1, self.strides[-2], self.strides[-1], 1)
-        self.add_prim_attr("strides", self.strides)
-
-
 class MaxPool3D(Primitive):
     r"""
     Applies a 3D max pooling over an input Tensor which can be regarded as a composition of 3D planes.
@@ -2026,19 +2011,21 @@ class MaxPool3D(Primitive):
         strides (Union[int, tuple[int]]): The distance of kernel moving, an int number that represents
             not only the depth, height of movement but also the width of movement,, or a tuple of three int numbers that
             represent depth, height and width of movement respectively. Default: ``1`` .
-        pad_mode (str): The optional value of pad mode is ``"SAME"`` , ``"VALID"`` or ``"PAD"`` .
-            Default: ``"VALID"`` .
-
-            - ``"SAME"``: Adopts the way of completion. The height and width of the output will be the same
-              as the input. The total number of padding will be calculated in horizontal and vertical
-              directions and evenly distributed to top, bottom, left and right if possible.
-              Otherwise, the last extra padding will be done from the bottom and the right side.
-
-            - ``"VALID"``: Adopts the way of discarding. The possible largest height and width of output
-              will be returned without padding. Extra pixels will be discarded.
-
-            - ``"PAD"``: Implicit paddings on both sides of the input in depth, height and width. The number of
-              ``"PAD"`` will be padded to the input Tensor borders. "pad_list" must be greater than or equal to 0.
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"SAME"`` , ``"VALID"`` or ``"PAD"`` . Default: ``"VALID"`` .
+
+            - ``"SAME"``: Pad the input around its depth/height/width dimension so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally.  If the amount is even,
+              it isuniformly distributed around the input, if it is odd, the excess amount goes
+              to the front/right/bottom side.
+              If this mode is set, `pad_list` must be 0.
+            - ``"VALID"``: No padding is applied to the input, and the output returns the maximum
+              possible depth, height and width. Extra pixels that could not complete a full stride will
+              be discarded. If this mode is set, `pad_list` must be 0.
+            - ``"PAD"``: Pad the input with a specified amount. In this mode, the amount of padding
+              in the depth, height and width dimension is determined by the `pad_list` parameter.
+              If this mode is set, `pad_list` must be greater than or equal to 0.
 
         pad_list (Union(int, tuple[int])): The pad value to be filled. Default: ``0`` . If `pad` is an integer, the
             paddings of head, tail, top, bottom, left and right are the same, equal to pad. If `pad` is a tuple of six
@@ -2347,14 +2334,17 @@ class AvgPool(Primitive):
         strides (Union[int, tuple[int]]): The distance of kernel moving, an int number that represents
             the height and width of movement are both strides, or a tuple of two int numbers that
             represent height and width of movement respectively. Default: ``1`` .
-        pad_mode (str, optional): The optional value for pad mode, is ``'same'`` or ``'valid'`` .
-            Default: ``'valid'`` .
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` or ``"valid"`` . Default: ``"valid"`` .
 
-            - ``'same'``: The height and width of the output are the same as the input divided by 'strides'
-              and rounded up.
+            - ``"same"``: Pad the input around its edges so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally, If the amount is even, it is
+              uniformly distributed around the input, if it is odd, the excess amount goes to the right/bottom side.
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible height and width. Extra pixels that could not complete a full stride will
+              be discarded.
 
-            - ``'valid'``: Returns the output of the valid calculation without filling. Redundant pixels that
-              do not satisfy the calculation will be discarded.
         data_format (str, optional): The format of input and output data. It should be ``'NHWC'`` or ``'NCHW'`` .
             Default: ``'NCHW'`` .
 
@@ -2451,16 +2441,17 @@ class AvgPoolV1(Primitive):
         strides (Union[int, tuple[int]]): The distance of kernel moving, an integer that represents
             the height and width of movement are both strides, or a tuple of two integers that
             represent height and width of movement, respectively. Default: ``1`` .
-        pad_mode (str): The optional value for pad mode, should be one of ``"same"`` or ``"valid"`` .
-            Default: ``"valid"`` .
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` or ``"valid"`` . Default: ``"valid"`` .
 
-            - ``"same"``: Adopts the way of completion. The height and width of output will be the same as
-              the input. The total number of padding will be calculated horizontally and vertically,
-              and evenly distributed to top and bottom, left and right if possible.
-              Otherwise, the last extra padding will be done from bottom and right.
+            - ``"same"``: Pad the input around its edges so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally, If the amount is even, it is
+              uniformly distributed around the input, if it is odd, the excess amount goes to the right/bottom side.
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible height and width. Extra pixels that could not complete a full stride will
+              be discarded.
 
-            - ``"valid"``: Adopts the way of discarding. The largest possible height and width of output
-              will be returned without padding. Extra pixels will be discarded.
         data_format (str): The format of input and output data. Should be ``'NHWC'`` or ``'NCHW'`` .
             Default: ``'NCHW'`` .
 
@@ -2708,8 +2699,21 @@ class Conv2DTranspose(Conv2DBackpropInput):
     Args:
         out_channel (int): The dimensionality of the output space.
         kernel_size (Union[int, tuple[int]]): The size of the convolution window.
-        pad_mode (str): Modes to fill padding. It could be ``"valid"`` , ``"same"`` , or ``"pad"`` .
-            Default: ``"valid"`` .
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` , ``"valid"`` or ``"pad"`` . Default: ``"valid"`` .
+
+            - ``"same"``: Pad the input around its edges so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally, If the amount is even, it is
+              uniformly distributed around the input, if it is odd, the excess amount goes to the right/bottom side.
+              If this mode is set, `pad` must be 0.
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible height and width. Extra pixels that could not complete a full stride will
+              be discarded. If this mode is set, `pad` must be 0.
+            - ``"pad"``: Pad the input with a specified amount. In this mode, the amount of padding
+              in the height and width directions is determined by the `pad` parameter.
+              If this mode is set, `pad` must be greater than or equal to 0.
+
             Please refer to :class:`mindspore.nn.Conv2dTranspose` for more specifications about `pad_mode`.
         pad (Union[int, tuple[int]]): The pad value to be filled. Default: ``0`` . If `pad` is an integer, the paddings
                     of top, bottom, left and right are the same, equal to pad. If `pad` is a tuple of four integers,
@@ -2779,9 +2783,13 @@ class BiasAdd(Primitive):
             Default is ``"NCHW"`` .
 
     Inputs:
-        - **input_x** (Tensor) - The input tensor. The shape can be 2-5 dimensions.
+        - **input_x** (Tensor) - The input tensor. The shape can be 2-5 dimensions. Supported dtypes:
+
+          - Ascend/CPU: all Number type.
+          - GPU: float16, float32, int8.
+
         - **bias** (Tensor) - The bias tensor, with shape :math:`(C)`. C must be the same as channel dimension C of
-          `input_x`.
+          `input_x`. It has the same type as `input_x`.
 
     Outputs:
         Tensor, with the same shape and data type as `input_x`.
@@ -2790,7 +2798,7 @@ class BiasAdd(Primitive):
         TypeError: If `data_format` is not a str.
         ValueError: If value of `data_format` is not in the range of ['NHWC','NCHW','NCDHW'].
         TypeError: If `input_x` or `bias` is not a Tensor.
-        TypeError: If dtype of `input_x` or `bias` is inconsistent.
+        TypeError: If dtype of `input_x` and `bias` is inconsistent.
         TypeError: If dimension of `input_x` is not in the range [2, 5].
 
     Supported Platforms:
@@ -2820,7 +2828,7 @@ class NLLLoss(Primitive):
     r"""
     Gets the negative log likelihood loss between logits and labels.
 
-    The nll loss with reduction=none can be described as:
+    The nll loss with :math:`reduction = none` can be described as:
 
     .. math::
 
@@ -2831,7 +2839,7 @@ class NLLLoss(Primitive):
     where :math:`x` is the logits, :math:`t` is the labels, :math:`w` is the weight,
     N is the batch size, :math:`c` belonging to [0, C-1] is class index, where :math:`C` is the number of classes.
 
-    If reduction is not ``'none'`` (default ``'mean'`` ), then
+    If :math:`reduction \neq none` (default ``'mean'`` ), then
 
     .. math::
 
@@ -2841,8 +2849,13 @@ class NLLLoss(Primitive):
         \end{array}\right.
 
     Args:
-        reduction (str): Apply specific reduction method to the output: ``"none"`` , ``"mean"`` , or ``"sum"`` .
-          Default: ``"mean"`` .
+        reduction (str, optional): Apply specific reduction method to the output: ``'none'`` , ``'mean'`` ,
+            ``'sum'`` . Default: ``'mean'`` .
+
+            - ``'none'``: no reduction will be applied.
+            - ``'mean'``: compute and return the weighted mean of elements in the output.
+            - ``'sum'``: the output elements will be summed.
+
         ignore_index (int): Specifies a target value that is ignored
             and does not contribute to the input gradient. Default: ``-100`` .
 
@@ -2856,8 +2869,9 @@ class NLLLoss(Primitive):
     Outputs:
         Tuple of 2 tensors composed with `loss` and `total_weight`.
 
-        - **loss** (Tensor) - When `reduction` is 'none' and `logits` is a 2D tensor, the `loss` shape is :math:`(N,)`.
-          Otherwise, the `loss` is a scalar. The data type is the same with `input's`.
+        - **loss** (Tensor) - When `reduction` is ``'none'`` and `logits` is a 2D tensor,
+          the `loss` shape is :math:`(N,)`. Otherwise, the `loss` is a scalar.
+          The data type is the same with `input's`.
         - **total_weight** (Tensor) - The `total_weight` is a scalar. The data type is the same with `weight's`.
 
     Raises:
@@ -3155,6 +3169,10 @@ class SmoothL1Loss(Primitive):
         reduction (str, optional): Apply specific reduction method to the output: ``'none'`` , ``'mean'`` ,
             ``'sum'`` . Default: ``'none'`` .
 
+            - ``'none'``: no reduction will be applied.
+            - ``'mean'``: compute and return the mean of elements in the output.
+            - ``'sum'``: the output elements will be summed.
+
     Inputs:
         - **logits** (Tensor) - Input Tensor of any dimension. Data type must be float16, float32 or float64.
         - **labels** (Tensor) - Ground truth data, has the same shape and dtype as the `logits`.
@@ -3202,12 +3220,12 @@ class MultiMarginLoss(Primitive):
     Args:
         p (int, optional): The norm degree for pairwise distance. Should be 1 or 2. Default: ``1`` .
         margin (int, optional): A parameter to change pairwise distance. Default: ``1.0`` .
-        reduction (str, optional): Apply specific reduction method to the output: ``"none"`` ,
-            ``"mean"`` , ``"sum"`` . Default: ``"mean"`` .
+        reduction (str, optional): Apply specific reduction method to the output: ``'none'`` , ``'mean'`` ,
+            ``'sum'`` . Default: ``'mean'`` .
 
-            - ``"none"``: no reduction will be applied.
-            - ``"mean"``: the sum of the output will be divided by the number of elements in the output.
-            - ``"sum"``: the output will be summed.
+            - ``'none'``: no reduction will be applied.
+            - ``'mean'``: compute and return the weighted mean of elements in the output.
+            - ``'sum'``: the output elements will be summed.
 
     Inputs:
         - **inputs** (Tensor) - Input , with shape :math:`(N, C)`. Data type only support float32, float16
@@ -3218,7 +3236,7 @@ class MultiMarginLoss(Primitive):
           support float16, float32 or float64.
 
     Outputs:
-        Tensor, When `reduction` is 'none', the shape is :math:`(N,)`.
+        Tensor, When `reduction` is ``'none'``, the shape is :math:`(N,)`.
         Otherwise, it is a scalar. Has the same data type with `inputs`.
 
     Supported Platforms:
@@ -3261,15 +3279,19 @@ class SoftMarginLoss(Primitive):
     where :math:`x.nelement()` is the number of elements of x.
 
     Args:
-        reduction (str): Apply specific reduction method to the output: ``"none"`` , ``"mean"`` or ``"sum"`` .
-          Default: ``"mean"`` .
+        reduction (str, optional): Apply specific reduction method to the output: ``'none'`` , ``'mean'`` ,
+            ``'sum'`` . Default: ``'mean'`` .
+
+            - ``'none'``: no reduction will be applied.
+            - ``'mean'``: compute and return the mean of elements in the output.
+            - ``'sum'``: the output elements will be summed.
 
     Inputs:
         - **logits** (Tensor) - Predict data. Data type must be float16 or float32.
         - **labels** (Tensor) - Ground truth data, with the same type and shape as `logits`.
 
     Outputs:
-        Tensor or Scalar, if `reduction` is "none", its shape is the same as `logits`.
+        Tensor or Scalar, if `reduction` is ``"none"``, its shape is the same as `logits`.
         Otherwise, a scalar value will be returned.
 
     Raises:
@@ -3736,26 +3758,28 @@ class LayerNorm(Primitive):
 
     Args:
         begin_norm_axis (int): The begin axis of the `input_x` to apply LayerNorm,
-            the value must be in [-1, rank(input)). Default: ``1`` .
+            the value must be in [-1, rank(input_x)). Default: ``1`` .
         begin_params_axis (int): The begin axis of the parameter input (`gamma`, `beta`) to
-            apply LayerNorm, the value must be in [-1, rank(input)). Default: ``1`` .
-        epsilon (float): A value added to the denominator for numerical stability. Default: ``1e-7`` .
+            apply LayerNorm, the value must be in [-1, rank(input_x)). Default: ``1`` .
+        epsilon (float): A value added to the denominator for numerical stability(:math:`\epsilon`). Default: ``1e-7`` .
 
     Inputs:
         - **input_x** (Tensor) - Tensor of shape :math:`(N, \ldots)`.
           The input of LayerNorm. Supported dtypes: float16, float32, float64.
-        - **gamma** (Tensor) - Tensor of shape :math:`(P_0, \ldots, P_\text{begin_params_axis})`.
+        - **gamma** (Tensor) - Tensor of shape :math:`(P_\text{begin_params_axis}, \ldots, P_\text{rank(input_x)-1})`.
           The learnable parameter :math:`\gamma` as the scale on norm. Supported dtypes: float16, float32, float64.
-        - **beta** (Tensor) - Tensor of shape :math:`(P_0, \ldots, P_\text{begin_params_axis})`.
+        - **beta** (Tensor) - Tensor of shape :math:`(P_\text{begin_params_axis}, \ldots, P_\text{rank(input_x)-1})`.
           The learnable parameter :math:`\beta` as the scale on norm. Supported dtypes: float16, float32, float64.
 
     Outputs:
         tuple[Tensor], tuple of 3 tensors, the normalized input and the updated parameters.
 
         - **output_x** (Tensor) - The normalized input, has the same type and shape as the `input_x`.
-          The shape is :math:`(N, C)`.
-        - **mean** (Tensor) - Tensor of shape :math:`(C,)`.
-        - **variance** (Tensor) - Tensor of shape :math:`(C,)`.
+        - **mean** (Tensor) - The first `begin_norm_axis` dimensions of `mean` shape is the same as `input_x`,
+          and the remaining dimensions are 1. Suppose the shape of the `input_x` is :math:`(x_1, x_2, \ldots, x_R)`,
+          the shape of the `mean` is :math:`(x_1, \ldots, x_{begin_params_axis}, 1, \ldots, 1)`
+          (when `begin_params_axis=0`, the shape of `mean` is :math:`(1, \ldots, 1)` ).
+        - **variance** (Tensor) - Shape is the same as `mean` .
 
     Raises:
         TypeError: If `begin_norm_axis` or `begin_params_axis` is not an int.
@@ -3855,38 +3879,6 @@ class L2Normalize(Primitive):
         self.axis = axis
 
 
-class DropoutGenMask(Primitive):
-    """
-    The DropoutGenMask interface is deprecated, please use the :class:`mindspore.ops.Dropout` instead.
-
-    Supported Platforms:
-        Deprecated
-    """
-
-    @deprecated("1.5", "ops.Dropout", False)
-    @prim_attr_register
-    def __init__(self, Seed0=0, Seed1=0):
-        """Initialize DropoutGenMask."""
-        self.init_prim_io_names(inputs=['shape', 'keep_prob'], outputs=['output'])
-        validator.check_value_type("Seed0", Seed0, [int], self.name)
-        validator.check_value_type("Seed1", Seed1, [int], self.name)
-        self.add_prim_attr("side_effect_hidden", True)
-
-
-class DropoutDoMask(Primitive):
-    """
-    The DropoutDoMask interface is deprecated, please use the :class:`mindspore.ops.Dropout` instead.
-
-    Supported Platforms:
-        Deprecated
-    """
-
-    @deprecated("1.5", "ops.Dropout", False)
-    @prim_attr_register
-    def __init__(self):
-        pass
-
-
 class ResizeBilinear(PrimitiveWithInfer):
     r"""
     This API is deprecated, please use the :class:`mindspore.ops.ResizeBilinearV2` instead.
@@ -4017,6 +4009,7 @@ class OneHot(Primitive):
 
     Note:
         If the input indices is rank `N`, the output will have rank `N+1`. The new axis is created at dimension `axis`.
+        On Ascend, if `on_value` is Int64 dtype, `indices` must be Int64 dtype.
 
     Args:
         axis (int): Position to insert the value. e.g. If shape of `indices` is :math:`(N, C)`, and `axis` is -1,
@@ -4025,18 +4018,20 @@ class OneHot(Primitive):
 
     Inputs:
         - **indices** (Tensor) - A tensor of indices. Tensor of shape :math:`(X_0, \ldots, X_n)`.
-          Data type must be uint8, int32 or int64.
+          Data type must be int32 or int64.
         - **depth** (int) - A scalar defining the depth of the one-hot dimension.
-        - **on_value** (Tensor) - A value to fill in output when `indices[j] = i`.
+        - **on_value** (Tensor) - A value to fill in output when `indices[j] = i`. Data type must be int32, int64,
+          float16 or float32.
         - **off_value** (Tensor) - A value to fill in output when `indices[j] != i`.
           It has the same data type as `on_value`.
 
     Outputs:
-        Tensor, one-hot tensor. Tensor of shape :math:`(X_0, \ldots, X_{axis}, \text{depth} ,X_{axis+1}, \ldots, X_n)`.
+        Tensor, one-hot tensor. Tensor of shape :math:`(X_0, \ldots, X_{axis}, \text{depth} ,X_{axis+1}, \ldots, X_n)`,
+        and it has the same data type as `on_value`.
 
     Raises:
         TypeError: If `axis` or `depth` is not an int.
-        TypeError: If dtype of `indices` is not uint8, int32 or int64.
+        TypeError: If dtype of `indices` is not int32 or int64.
         TypeError: If `indices`, `on_value` or `off_value` is not a Tensor.
         ValueError: If `axis` is not in range [-1, len(indices_shape)].
         ValueError: If `depth` is less than 0.
@@ -4065,26 +4060,6 @@ class OneHot(Primitive):
         validator.check_value_type("axis", axis, [int], self.name)
 
 
-class Gelu(PrimitiveWithInfer):
-    """
-    Same as operator GeLU. Gelu will be deprecated in the future.
-    Please use GeLU instead.
-    """
-
-    @deprecated("1.1", "GeLU", True)
-    @prim_attr_register
-    def __init__(self):
-        """Initialize Gelu"""
-        self.init_prim_io_names(inputs=['x'], outputs=['output'])
-
-    def infer_shape(self, input_x):
-        return input_x
-
-    def infer_dtype(self, input_x):
-        validator.check_tensor_dtype_valid("input_x", input_x, (mstype.float16, mstype.float32), self.name)
-        return input_x
-
-
 class GeLU(Primitive):
     r"""
     Gaussian Error Linear Units activation function.
@@ -4131,26 +4106,6 @@ class GeLU(Primitive):
         self.init_prim_io_names(inputs=['x'], outputs=['output'])
 
 
-class FastGelu(PrimitiveWithInfer):
-    """
-    Same as operator FastGeLU. FastGelu will be deprecated in the future.
-    Please use FastGeLU instead.
-    """
-
-    @deprecated("1.1", "FastGeLU", True)
-    @prim_attr_register
-    def __init__(self):
-        """Initialize FastGelu."""
-        self.init_prim_io_names(inputs=['x'], outputs=['output'])
-
-    def infer_shape(self, input_x):
-        return input_x
-
-    def infer_dtype(self, input_x):
-        validator.check_tensor_dtype_valid("input_x", input_x, (mstype.float16, mstype.float32), self.name)
-        return input_x
-
-
 class FastGeLU(Primitive):
     r"""
     Fast Gaussian Error Linear Units activation function.
@@ -4301,19 +4256,24 @@ class LSTM(Primitive):
         bidirectional (bool): Specifies whether it is a bidirectional LSTM.
         dropout (float): If not 0, append `Dropout` layer on the outputs of each
             LSTM layer except the last layer. The range of dropout is [0.0, 1.0].
+        proj_size (int): If `proj_size` > 0, a projection of the corresponding size will be used,
+            which is only supported on CPU now. Default: ``0`` .
 
     Inputs:
         - **input** (Tensor) - Tensor of shape :math:`(seq\_len, batch\_size, input\_size)` or
           :math:`(batch\_size, seq\_len, input\_size)`.
-        - **h** (Tensor) - Tensor of shape :math:`(num\_directions * num\_layers, batch\_size, hidden\_size)`.
+        - **h** (Tensor) - Tensor of shape :math:`(num\_directions * num\_layers, batch\_size, real\_hidden\_size)`.
         - **c** (Tensor) - Tensor of shape :math:`(num\_directions * num\_layers, batch\_size, hidden\_size)`.
         - **w** (Tensor) - A weight Tensor.
 
+        If :math:`proj\_size > 0` , :math:`real\_hidden\_size = proj\_size` , otherwise
+        :math:`real\_hidden\_size = hidden\_size` .
+
     Outputs:
-        Tuple, a tuple contains (`output`, `h_n`, `c_n`, `reserve`, `state`).
+        Tuple, a tuple contains `(output, h_n, c_n, reserve, state)`.
 
-        - **output** (Tensor) - Tensor of shape :math:`(seq\_len, batch\_size, num\_directions * hidden\_size)`.
-        - **h_n** (Tensor) - Tensor of shape :math:`(num\_directions * num\_layers, batch\_size, hidden\_size)`.
+        - **output** (Tensor) - Tensor of shape :math:`(seq\_len, batch\_size, num\_directions * real\_hidden\_size)`.
+        - **h_n** (Tensor) - Tensor of shape :math:`(num\_directions * num\_layers, batch\_size, real\_hidden\_size)`.
         - **c_n** (Tensor) - Tensor of shape :math:`(num\_directions * num\_layers, batch\_size, hidden\_size)`.
         - **reserve** (Tensor) - Tensor of shape :math:`(r, 1)`.
         - **state** (Tensor) - Random number generator state and its shape is :math:`(s, 1)`.
@@ -4323,6 +4283,7 @@ class LSTM(Primitive):
         TypeError: If `has_bias` or `bidirectional` is not a bool.
         TypeError: If `dropout` is not a float.
         ValueError: If `dropout` is not in range [0.0, 1.0].
+        ValueError: If `proj_size` is not in range [0, `hidden_size`).
 
     Supported Platforms:
         ``GPU`` ``CPU``
@@ -4356,10 +4317,12 @@ class LSTM(Primitive):
     """
 
     @prim_attr_register
-    def __init__(self, input_size, hidden_size, num_layers, has_bias, bidirectional, dropout):
+    def __init__(self, input_size, hidden_size, num_layers, has_bias, bidirectional, dropout, proj_size=0):
         """Initialize LSTM."""
         self.input_size = validator.check_positive_int(input_size, "input_size", self.name)
         self.hidden_size = validator.check_positive_int(hidden_size, "hidden_size", self.name)
+        self.proj_size = validator.check_int_range(proj_size, 0, hidden_size, validator.INC_LEFT,
+                                                   'proj_size', self.name)
         self.num_layers = validator.check_positive_int(num_layers, "num_layers", self.name)
         self.has_bias = validator.check_value_type("has_bias", has_bias, (bool,), self.name)
         self.bidirectional = validator.check_value_type("bidirectional", bidirectional, (bool,), self.name)
@@ -4466,8 +4429,12 @@ class BCEWithLogitsLoss(PrimitiveWithInfer):
     :math:`P_c>1` increases the recall, :math:`P_c<1` increases the precision.
 
     Args:
-        reduction (str): Type of reduction to be applied to loss. The optional values are ``'mean'`` , ``'sum'`` , and
-            ``'none'`` , not case sensitive. If ``'none'`` , do not perform reduction. Default: ``'mean'`` .
+        reduction (str, optional): Apply specific reduction method to the output: ``'none'`` , ``'mean'`` ,
+            ``'sum'`` . Default: ``'mean'`` .
+
+            - ``'none'``: no reduction will be applied.
+            - ``'mean'``: compute and return the weighted mean of elements in the output.
+            - ``'sum'``: the output elements will be summed.
 
     Inputs:
         - **logits** (Tensor) - Input logits. Data type must be float16 or float32.
@@ -4481,7 +4448,7 @@ class BCEWithLogitsLoss(PrimitiveWithInfer):
           Data type must be float16 or float32.
 
     Outputs:
-        Tensor or Scalar, if `reduction` is 'none', it's a tensor with the same shape and type as input `logits`.
+        Tensor or Scalar, if `reduction` is ``'none'``, it's a tensor with the same shape and type as input `logits`.
         Otherwise, the output is a scalar.
 
     Raises:
@@ -4489,7 +4456,7 @@ class BCEWithLogitsLoss(PrimitiveWithInfer):
         TypeError: If data type of any input is neither float16 nor float32.
         TypeError: If data type of `reduction` is not string.
         ValueError: If `weight` or `pos_weight` can not be broadcast to a tensor with shape of `logits`.
-        ValueError: If `reduction` is not one of 'none', 'mean' or 'sum'.
+        ValueError: If `reduction` is not one of ``'none'``, ``'mean'`` or ``'sum'``.
 
     Supported Platforms:
         ``Ascend`` ``GPU`` ``CPU``
@@ -4669,9 +4636,15 @@ class MirrorPad(Primitive):
     Pads the input tensor according to the paddings and mode.
 
     Args:
-        mode (str): Specifies the padding mode. The optional values are ``'REFLECT'`` and ``'SYMMETRIC'`` .
+        mode (str, optional): An optional string specifying the pad method.
+            The optional values are ``'REFLECT'`` and ``'SYMMETRIC'`` .
             Default: ``'REFLECT'`` .
 
+            - ``'REFLECT'``: Reflect the value on the edge while omitting the last one.
+              For example, pad [1, 2, 3, 4] with 2 elements on both sides will result in [3, 2, 1, 2, 3, 4, 3, 2].
+            - ``'SYMMETRIC'``: Reflect the value on the edge while repeating the last one.
+              For example, pad [1, 2, 3, 4] with 2 elements on both sides will result in [2, 1, 1, 2, 3, 4, 4, 3].
+
     Inputs:
         - **input_x** (Tensor) - Tensor of shape :math:`(N, *)`, where :math:`*` means, any number of
           additional dimensions.
@@ -4683,15 +4656,14 @@ class MirrorPad(Primitive):
           paddings[D, 0] and paddings[D, 1] must be no greater than input_x.dim_size(D)
           (or input_x.dim_size(D) - 1) if mode is SYMMETRIC (if REFLECT, respectively).
 
-
     Outputs:
         Tensor, the tensor after padding.
 
-        - If `mode` is "REFLECT", it uses a way of symmetrical copying through the axis of symmetry to fill in.
+        - If `mode` is ``'REFLECT'``, it uses a way of symmetrical copying through the axis of symmetry to fill in.
           If the `input_x` is [[1,2,3], [4,5,6], [7,8,9]] and `paddings` is [[1,1], [2,2]], then the
           `Outputs` is [[6,5,4,5,6,5,4], [3,2,1,2,3,2,1], [6,5,4,5,6,5,4], [9,8,7,8,9,8,7], [6,5,4,5,6,5,4]].
           For a more intuitive understanding, please see the example below.
-        - If `mode` is "SYMMETRIC", the filling method is similar to the "REFLECT". It is also copied
+        - If `mode` is ``'SYMMETRIC'``, the filling method is similar to the ``'REFLECT'``. It is also copied
           according to the symmetry axis, except that it includes the symmetry axis. If the `input_x`
           is [[1,2,3], [4,5,6], [7,8,9]] and `paddings` is [[1,1], [2,2]], then the `Outputs` is
           [[2,1,1,2,3,3,2], [2,1,1,2,3,3,2], [5,4,4,5,6,6,5], [8,7,7,8,9,9,8], [8,7,7,8,9,9,8]].
@@ -5675,7 +5647,7 @@ class KLDivLoss(Primitive):
         - **labels** (Tensor) - The label Tensor which has the same shape and data type as `logits`.
 
     Outputs:
-        Tensor or Scalar, if `reduction` is 'none', then output is a tensor and has the same shape as `logits`.
+        Tensor or Scalar, if `reduction` is ``'none'``, then output is a tensor and has the same shape as `logits`.
         Otherwise it is a scalar.
 
     Raises:
@@ -5750,8 +5722,12 @@ class BinaryCrossEntropy(Primitive):
         - The value of :math:`x` must range from 0 to 1.
 
     Args:
-        reduction (str): Specifies the reduction to be applied to the output.
-            Its value must be one of ``'none'`` , ``'mean'`` or ``'sum'`` . Default: ``'mean'`` .
+        reduction (str, optional): Apply specific reduction method to the output: ``'none'`` , ``'mean'`` ,
+            ``'sum'`` . Default: ``'mean'`` .
+
+            - ``'none'``: no reduction will be applied.
+            - ``'mean'``: compute and return the weighted mean of elements in the output.
+            - ``'sum'``: the output elements will be summed.
 
     Inputs:
         - **logits** (Tensor) - The predictive value whose data type must be float16 or float32,
@@ -5766,7 +5742,7 @@ class BinaryCrossEntropy(Primitive):
 
     Raises:
         TypeError: If dtype of `logits`, `labels` or `weight` (if given) is neither float16 nor float32.
-        ValueError: If `reduction` is not one of 'none', 'mean' or 'sum'.
+        ValueError: If `reduction` is not one of ``'none'``, ``'mean'`` or ``'sum'``.
         ValueError: If shape of `labels` is not the same as `logits` or `weight` (if given).
         TypeError: If `logits`, `labels` or `weight` is not a Tensor.
 
@@ -7173,7 +7149,19 @@ class Dropout(PrimitiveWithCheck):
 
     Outputs:
         - **output** (Tensor) - With the same shape and data type as `x`.
-        - **mask** (Tensor) - With the same shape as `x`.
+        - **mask** (Tensor) - The mask applied to `x`.
+
+          - On GPU and CPU, `mask` has the same shape and data type as `x`.
+          - On Ascend, to achieve a better performance, it is denoted as a 1-D Tensor
+            with Uint8 data type. It has shape :math:`(byte\_counts, )` where :math:`byte\_counts` is the
+            number of bytes needed to mask the input `x`, :math:`byte\_counts` is calculated using the
+            following formula:
+
+            .. math::
+
+                byte\_counts = \text{ceil}(\text{cumprod}(x.shape) / 128) * 16
+
+            If shape of `x` is :math:`(2, 3, 4, 5, 6)`, the shape of `mask` will be :math:`(96, )`.
 
     Supported Platforms:
         ``Ascend`` ``GPU`` ``CPU``
@@ -7195,6 +7183,7 @@ class Dropout(PrimitiveWithCheck):
         self.seed0 = validator.check_value_type("Seed0", Seed0, [int], self.name)
         self.seed1 = validator.check_value_type("Seed1", Seed1, [int], self.name)
         self.keep_prob = validator.check_float_range(keep_prob, 0, 1, validator.INC_RIGHT, "keep_prob", self.name)
+        self.add_prim_attr("side_effect_hidden", True)
 
     def check_shape(self, x_shape):
         validator.check_int(len(x_shape), 1, validator.GE, "x_shape", self.name)
@@ -7402,6 +7391,9 @@ class CTCGreedyDecoder(Primitive):
 
     Refer to :func:`mindspore.ops.ctc_greedy_decoder` for more details.
 
+    Note:
+        On Ascend, 'merge_repeated' can not be set to false.
+
     Args:
         merge_repeated (bool, optional): If ``True`` , merge repeated classes in output. Default: ``True`` .
 
@@ -7824,6 +7816,10 @@ class LRN(Primitive):
     r"""
     Local Response Normalization.
 
+    .. warning::
+        LRN is deprecated on Ascend due to potential accuracy problem. It's recommended to use other
+        normalization methods, e.g. :class:`mindspore.ops.BatchNorm`.
+
     .. math::
 
         b_{c} = a_{c}\left(k + \frac{\alpha}{n}
@@ -7854,7 +7850,7 @@ class LRN(Primitive):
         TypeError: If `x` is not a Tensor.
 
     Supported Platforms:
-        ``Ascend`` ``GPU`` ``CPU``
+        ``GPU`` ``CPU``
 
     Examples:
         >>> import mindspore
@@ -7908,21 +7904,22 @@ class AvgPool3D(Primitive):
         strides (Union[int, tuple[int]]): The distance of kernel moving, an int number that represents
             the depth, height and width of movement are both strides, or a tuple of three int numbers that
             represent depth, height and width of movement respectively. Default: ``1`` .
-        pad_mode (str): The optional value for pad mode, is ``"same"`` , ``"valid"`` , ``"pad"`` .
-            Default: ``"valid"`` .
-
-            - ``"same"``: Adopts the way of completion. The depth, height and width of the output will be the same
-              as the input. The total number of padding will be calculated in depth, horizontal and vertical
-              directions and evenly distributed to head and tail, top and bottom, left and right if possible.
-              Otherwise, the last extra padding will be done from the tail, bottom and the right side.
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` , ``"valid"`` or ``"pad"`` . Default: ``"valid"`` .
+
+            - ``"same"``: Pad the input around its depth/height/width dimension so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally.  If the amount is even,
+              it isuniformly distributed around the input, if it is odd, the excess amount goes
+              to the front/right/bottom side.
               If this mode is set, `pad` must be 0.
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible depth, height and width. Extra pixels that could not complete a full stride will
+              be discarded. If this mode is set, `pad` must be 0.
+            - ``"pad"``: Pad the input with a specified amount. In this mode, the amount of padding
+              in the depth, height and width dimension is determined by the `pad` parameter.
+              If this mode is set, `pad` must be greater than or equal to 0.
 
-            - ``"valid"``: Adopts the way of discarding. The possible largest depth, height and width of output
-              will be returned without padding. Extra pixels will be discarded. If this mode is set, `pad`
-              must be 0.
-
-            - pad: Implicit paddings on both sides of the input in depth, height, width. The number of `pad` will
-              be padded to the input Tensor borders. `pad` must be greater than or equal to 0.
         pad (Union(int, tuple[int], list[int])): The pad value to be filled. Default: ``0`` . If `pad` is an integer,
             the paddings of head, tail, top, bottom, left and right are the same, equal to pad.
             If `pad` is a tuple of six integers, the padding of head, tail, top, bottom, left and right equal to
@@ -8005,74 +8002,97 @@ class AvgPool3D(Primitive):
 
 class Conv3D(Primitive):
     r"""
-    Applies a 3D convolution over an input tensor. The input tensor is typically of shape
-    :math:`(N, C_{in}, D_{in}, H_{in}, W_{in})` and output shape
-    :math:`(N, C_{out}, D_{out}, H_{out}, W_{out})`, where :math:`N` is batch size, :math:`C` is channel number,
-    :math:`D` is depth, :math:`H, W` is feature height and width respectively.
-    the output value of a layer is calculated as:
+    3D convolution layer.
+
+    Applies a 3D convolution over an input tensor which is typically of shape
+    :math:`(N, C_{in}, D_{in}, H_{in}, W_{in})`,
+    where :math:`N` is batch size, :math:`C` is channel number, :math:`D` is feature depth,
+    :math:`H` is feature height, :math:`W` is feature width.
+
+    The output is calculated based on formula:
 
     .. math::
-        \operatorname{out}\left(N_{i}, C_{\text {out}_j}\right)=\operatorname{bias}\left(C_{\text {out}_j}\right)+
-        \sum_{k=0}^{C_{in}-1} ccor(\text {weight}\left(C_{\text {out}_j}, k\right),
-        \operatorname{input}\left(N_{i}, k\right))
-
-    where :math:`k` is kernel,
-    :math:`ccor` is the `cross-correlation <https://en.wikipedia.org/wiki/Cross-correlation>`_ ,
-    :math:`C_{in}` is the channel number of the input, :math:`out_{j}` corresponds to the :math:`j`-th channel of
-    the output and :math:`j` is in the range of :math:`[0, C_{out} - 1]`. :math:`\text{weight}(C_{\text{out}_j}, k)`
-    is a convolution kernel slice with shape
-    :math:`(\text{kernel_size[0]}, \text{kernel_size[1]}, \text{kernel_size[2]})`,
-    where :math:`\text{kernel_size[0]}`, :math:`\text{kernel_size[1]}` and :math:`\text{kernel_size[2]}` are
-    the depth, height and width of the convolution kernel respectively. :math:`\text{bias}` is the bias parameter
-    and :math:`\text{X}` is the input tensor.
-    The shape of full convolution kernel is
-    :math:`(C_{out}, C_{in} / \text{groups}, \text{kernel_size[0]}, \text{kernel_size[1]}, \text{kernel_size[2]})`,
-    where `groups` is the number of groups to split `input` in the channel dimension.
-
-    For more details, please refer to the paper `Gradient Based Learning Applied to Document
-    Recognition <http://vision.stanford.edu/cs598_spring07/papers/Lecun98.pdf>`_ .
+
+        \text{out}(N_i, C_{\text{out}_j}) = \text{bias}(C_{\text{out}_j}) +
+        \sum_{k = 0}^{C_{in} - 1} \text{ccor}({\text{weight}(C_{\text{out}_j}, k), \text{X}(N_i, k)})
+
+    where :math:`bias` is the output channel bias, :math:`ccor` is
+    the `cross-correlation <https://en.wikipedia.org/wiki/Cross-correlation>`_,
+    , :math:`weight` is the convolution kernel value and :math:`X` represents the input feature map.
+
+    Here are the indices' meanings:
+    - :math:`i` corresponds to the batch number, ranging from 0 to N-1, where N is the batch size of the input.
+
+    - :math:`j` corresponds to the output channel, ranging from 0 to C_{out}-1, where C_{out} is the number of
+      output channels, which is also equal to the number of kernels.
+
+    - :math:`k` corresponds to the input channel, ranging from 0 to C_{in}-1, where C_{in} is the number of
+      input channels, which is also equal to the number of channels in the convolutional kernels.
+
+    Therefore, in the above formula, :math:`{bias}(C_{out_j})` represents the bias of the :math:`j`-th
+    output channel, :math:`{weight}(C_{out_j}, k)` represents the slice of the :math:`j`-th convolutional
+    kernel in the :math:`k`-th channel, and :math:`{X}(N_i, k)` represents the slice of the :math:`k`-th input
+    channel in the :math:`i`-th batch of the input feature map.
+
+    The shape of the convolutional kernel is given by
+    :math:`(\text{kernel_size[0]}, \text{kernel_size[1]}, \text{kernel_size[2]})`
+    where :math:`kernel\_size[0]` , :math:`kernel\_size[1]` and :math:`kernel\_size[2]` are the depth,
+    height and width of the kernel, respectively.
+    If we consider the input and output channels as well as the `group` parameter, the complete kernel shape
+    will be :math:`(C_{out}, C_{in} / \text{group}, \text{kernel_size[0]},
+    \text{kernel_size[1]}, \text{kernel_size[2]})`,
+    where `group` is the number of groups dividing `x`'s input channel when applying group convolution.
+
+    For more details about convolution layer, please refer to `Gradient Based Learning Applied to Document Recognition
+    <http://vision.stanford.edu/cs598_spring07/papers/Lecun98.pdf>`_.
 
     Note:
-        On Ascend platform, `group = 1` must be satisfied.
+        1. On Ascend platform, `groups = 1` must be satisfied.
+        2. On Ascend `dilation` on depth only supports the case of 1.
 
     Args:
-        out_channel (int): The number of output channel :math:`C_{out}`.
-        kernel_size (Union[int, tuple[int]]): Specifies the depth, height
-            and width of the 3D convolution window. It can be a single int or a tuple of 3 integers.
-            Single int means the value is for the depth, height and width
-            of the kernel. A tuple of 3 ints corresponds to the depth, height and width of the kernel respectively.
+        out_channel (int): Specifies output channel :math:`C_{out}`.
+        kernel_size (Union[int, tuple[int]]): Specifies the depth, height and width of the 3D convolution kernel.
+            It can be a single int or a tuple of 3 integers. A single int means the value is for depth, height
+            and the width. A tuple of 3 ints means the first value is for depth and
+            the rest is for the height and width.
         mode (int, optional): Modes for different convolutions. It is currently not used. Default: ``1`` .
         stride (Union[int, tuple[int]], optional): The distance of kernel moving, it can be an int number
             that represents the depth, height and width of movement or a tuple of three int numbers that
             represent depth, height and width movement respectively. Default: ``1`` .
-        pad_mode (str, optional): Specifies padding mode. The optional values are
-            ``"same"`` , ``"valid"`` and ``"pad"`` . Default: ``"valid"`` .
-
-            - ``"same"``: Adopts the way of completion. The depth, height and width of the output will be equal to
-              the input `x` divided by stride. The padding will be evenly calculated in head and tail, top and bottom,
-              left and right directions possiblily.
-              Otherwise, the last extra padding will be calculated from the tail, bottom and the right side.
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` , ``"valid"`` or ``"pad"`` . Default: ``"valid"`` .
+
+            - ``"same"``: Pad the input around its depth/height/width dimension so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally.  If the amount is even,
+              it isuniformly distributed around the input, if it is odd, the excess amount goes
+              to the front/right/bottom side.
               If this mode is set, `pad` must be 0.
-
-            - ``"valid"``: Adopts the way of discarding. The possible largest depth, height and width of output
-              will be returned without padding. Extra pixels will be discarded. If this mode is set, `pad`
-              must be 0.
-
-            - ``"pad"``: Implicit paddings on both sides of the input in depth, height and width. The number of `pad`
-              will be padded to the input Tensor borders. `pad` must be greater than or equal to 0.
-
-        pad (Union(int, tuple[int]), optional): The pad value to be filled. Default: ``0`` .
-            If `pad` is an integer, the paddings
-            of head, tail, top, bottom, left and right are the same, equal to pad. If `pad` is a tuple of six
-            integers, the padding of head, tail, top, bottom, left and right equal to pad[0], pad[1], pad[2],
-            pad[3], pad[4] and pad[5] correspondingly.
-        dilation (Union[int, tuple[int]], optional): The data type is int or a tuple of 3 integers
-            :math:`(dilation_d, dilation_h, dilation_w)`. Currently, dilation on depth only supports the case of 1
-            on Ascend backend. Specifies the dilation rate to use for dilated convolution. If set :math:`k > 1`,
-            there will be :math:`k - 1` pixels skipped for each sampling location.
-            The value ranges for the depth, height, and width dimensions are [1, D], [1, H], and [1, W],
-            respectively. Default: ``1`` .
-        group (int, optional):The number of groups into which the filter is divided. `in_channels`
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible depth, height and width. Extra pixels that could not complete a full stride will
+              be discarded. If this mode is set, `pad` must be 0.
+            - ``"pad"``: Pad the input with a specified amount. In this mode, the amount of padding
+              in the depth, height and width dimension is determined by the `pad` parameter.
+              If this mode is set, `pad` must be greater than or equal to 0.
+
+        pad (Union(int, tuple[int]), optional): Specifies the amount of padding to apply on input
+            when `pad_mode` is set to ``"pad"``. It can be a single int or a tuple of 6 ints.
+            If `pad` is one integer, the paddings of head, tail, top, bottom,
+            left and right are the same, equal to `pad`. If `pad` is a tuple with 6 integers, the
+            paddings of head, tail, top, bottom, left and right is equal to pad[0],
+            pad[1], pad[2], pad[3], pad[4] and pad[5] accordingly. Default: ``0`` .
+        dilation (Union[int, tuple[int]], optional): Specifies the dilation rate to use for dilated convolution.
+            It can be a single int or a tuple of 3 integers. A single int means the dilation size is the same
+            in the depth, height and width directions. A tuple of 3 ints represents the dilation size in
+            the depth, height and width directions, respectively.
+            Assuming :math:`dilation=(d0, d1, d2)`, the convolutional kernel samples the input with a
+            spacing of :math:`d0-1` elements in the depth direction,
+            :math:`d1-1` elements in the height direction, :math:`d2-1` elements in the
+            width direction respectively. The values in the depth, height and width dimensions are in the
+            ranges [1, D], [1, H] and [1, W], respectively.
+            Default: ``1`` .
+        group (int, optional): The number of groups into which the filter is divided. `in_channels`
             and `out_channels` must be divisible by `group`. Default: ``1`` .
         data_format (str, optional): The optional value for data format. Currently only support ``"NCDHW"`` .
 
@@ -8088,7 +8108,7 @@ class Conv3D(Primitive):
     Outputs:
         Tensor, the value that applied 3D convolution. The shape is :math:`(N, C_{out}, D_{out}, H_{out}, W_{out})`.
 
-        `pad_mode` is 'same':
+        `pad_mode` is ``"same"``:
 
         .. math::
             \begin{array}{ll} \\
@@ -8097,7 +8117,7 @@ class Conv3D(Primitive):
                 W_{out} = \left \lceil{\frac{W_{in}}{\text{stride[2]}}} \right \rceil \\
             \end{array}
 
-        `pad_mode` is 'valid':
+        `pad_mode` is ``"valid"``:
 
         .. math::
             \begin{array}{ll} \\
@@ -8109,15 +8129,15 @@ class Conv3D(Primitive):
                 {\text{stride[2]}} + 1} \right \rfloor \\
             \end{array}
 
-        `pad_mode` is 'pad':
+        `pad_mode` is ``"pad"``:
 
         .. math::
             \begin{array}{ll} \\
-                D_{out} = \left \lfloor{\frac{D_{in} + padding[0] + padding[1] - (\text{dilation[0]} - 1) \times
+                D_{out} = \left \lfloor{\frac{D_{in} + pad[0] + pad[1] - (\text{dilation[0]} - 1) \times
                 \text{kernel_size[0]} - 1 }{\text{stride[0]}} + 1} \right \rfloor \\
-                H_{out} = \left \lfloor{\frac{H_{in} + padding[2] + padding[3] - (\text{dilation[1]} - 1) \times
+                H_{out} = \left \lfloor{\frac{H_{in} + pad[2] + pad[3] - (\text{dilation[1]} - 1) \times
                 \text{kernel_size[1]} - 1 }{\text{stride[1]}} + 1} \right \rfloor \\
-                W_{out} = \left \lfloor{\frac{W_{in} + padding[4] + padding[5] - (\text{dilation[2]} - 1) \times
+                W_{out} = \left \lfloor{\frac{W_{in} + pad[4] + pad[5] - (\text{dilation[2]} - 1) \times
                 \text{kernel_size[2]} - 1 }{\text{stride[2]}} + 1} \right \rfloor \\
             \end{array}
 
@@ -8138,12 +8158,56 @@ class Conv3D(Primitive):
         >>> import mindspore
         >>> import numpy as np
         >>> from mindspore import Tensor, ops
+        >>> # case 1: specify kernel_size with tuple, all parameters use default values.
         >>> x = Tensor(np.ones([16, 3, 10, 32, 32]), mindspore.float16)
         >>> weight = Tensor(np.ones([32, 3, 4, 3, 3]), mindspore.float16)
         >>> conv3d = ops.Conv3D(out_channel=32, kernel_size=(4, 3, 3))
         >>> output = conv3d(x, weight)
         >>> print(output.shape)
         (16, 32, 7, 30, 30)
+        >>> # case 2: specify kernel_size with int, all parameters use default values.
+        >>> x = Tensor(np.ones([10, 20, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([40, 20, 3, 3, 3]), mindspore.float32)
+        >>> conv3d = ops.Conv3D(out_channel=40, kernel_size=3)
+        >>> output = conv3d(x, weight)
+        >>> print(output.shape)
+        (10, 40, 30, 30, 30)
+         >>> # case 3: stride=(1, 2, 3), other parameters being default.
+        >>> x = Tensor(np.ones([10, 20, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([40, 20, 3, 3, 3]), mindspore.float32)
+        >>> conv3d = ops.Conv3D(out_channel=40, kernel_size=3, stride=(1, 2, 3))
+        >>> output = conv3d(x, weight)
+        >>> print(output.shape)
+        (10, 40, 30, 15, 10)
+         >>> # case 4: pad_mode="pad", other parameters being default.
+        >>> x = Tensor(np.ones([10, 20, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([40, 20, 3, 3, 3]), mindspore.float32)
+        >>> conv3d = ops.Conv3D(out_channel=40, kernel_size=3, pad_mode="pad", pad=2)
+        >>> output = conv3d(x, weight)
+        >>> print(output.shape)
+        (10, 40, 34, 34, 34)
+         >>> # case 5: dilation=(1, 1, 1), other parameters being default.
+        >>> x = Tensor(np.ones([10, 20, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([40, 20, 3, 3, 3]), mindspore.float32)
+        >>> conv3d = ops.Conv3D(out_channel=40, kernel_size=3, dilation=(1, 1, 1))
+        >>> output = conv3d(x, weight)
+        >>> print(output.shape)
+        (10, 40, 30, 30, 30)
+        >>> # case 6: group=1, other parameters being default.
+        >>> x = Tensor(np.ones([10, 20, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([40, 20, 3, 3, 3]), mindspore.float32)
+        >>> conv3d = ops.Conv3D(out_channel=40, kernel_size=3, group=1)
+        >>> output = conv3d(x, weight)
+        >>> print(output.shape)
+        (10, 40, 30, 30, 30)
+        >>> # case 7: All parameters are specified.
+        >>> x = Tensor(np.ones([10, 20, 32, 32, 32]), mindspore.float32)
+        >>> weight = Tensor(np.ones([40, 20, 3, 3, 3]), mindspore.float32)
+        >>> conv3d = ops.Conv3D(out_channel=40, kernel_size=3, stride=(1, 2, 3), pad_mode="pad",
+        ...                     pad=2, dilation=(1), group=1)
+        >>> output = conv3d(x, weight)
+        >>> print(output.shape)
+        (10, 40, 34, 17, 12)
     """
 
     @prim_attr_register
@@ -8218,8 +8282,22 @@ class Conv3DBackpropInput(Primitive):
         out_channel (int): The dimension of the output.
         kernel_size (Union[int, tuple[int]]): The kernel size of the 3D convolution.
         mode (int): Modes for different convolutions. Not currently used.
-        pad_mode (str): Modes to fill padding. It could be ``"valid"`` , ``"same"`` , or ``"pad"`` .
-                    Default: ``"valid"`` .
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` , ``"valid"`` or ``"pad"`` . Default: ``"valid"`` .
+
+            - ``"same"``: Pad the input around its depth/height/width dimension so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally.  If the amount is even,
+              it isuniformly distributed around the input, if it is odd, the excess amount goes
+              to the front/right/bottom side.
+              If this mode is set, `pad` must be 0.
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible depth, height and width. Extra pixels that could not complete a full stride will
+              be discarded. If this mode is set, `pad` must be 0.
+            - ``"pad"``: Pad the input with a specified amount. In this mode, the amount of padding
+              in the depth, height and width dimension is determined by the `pad` parameter.
+              If this mode is set, `pad` must be greater than or equal to 0.
+
         pad (Union(int, tuple[int])): The pad value to be filled. Default: ``0`` . If `pad` is an integer, the
                     paddings of head, tail, top, bottom, left and right are the same, equal to pad. If `pad` is a
                     tuple of four integers, the padding of head, tail, top, bottom, left and right equal to pad[0],
@@ -8443,13 +8521,14 @@ class CTCLossV2(Primitive):
 
     Args:
         blank (int, optional): The blank label. Default: ``0`` .
-        reduction (str, optional): Apply specific reduction method to the output. Currently only support ``'none'`` ,
-            not case sensitive. Default: ``"none"`` .
+        reduction (str, optional): Apply specific reduction method to the output. Currently only support ``'none'``.
+            Default: ``'none'`` .
+
         zero_infinity (bool, optional): If loss is infinite, this parameter determines whether to set that loss
             and its correlated gradient to zero. Default: ``False`` .
 
     Inputs:
-        - **log_probs** (Tensor) - A tensor of shape :math:`(T, C, N)`, where :math:`T` is input length, :math:`N` is
+        - **log_probs** (Tensor) - A tensor of shape :math:`(T, N, C)`, where :math:`T` is input length, :math:`N` is
           batch size and :math:`C` is number of classes (including blank). Supported dtypes: float32, float64.
         - **targets** (Tensor) - A tensor of shape :math:`(N, S)`, where :math:`S` is max target length,
           means the target sequences. Supported dtypes: int32, int64.
@@ -8601,35 +8680,37 @@ class Conv3DTranspose(Primitive):
             Single int means the value is for the depth, height and width of the kernel.
             A tuple of 3 ints means the first value is for the depth, the second value is for the height and the
             other is for the width of the kernel.
-        mode (int): Modes for different convolutions. Default is ``1`` . It is currently not used.
-        pad_mode (str): Specifies padding mode. The optional values are
-            ``"same"`` , ``"valid"`` , ``"pad"`` . Default: ``"valid"`` .
-
-            - ``"same"``: Adopts the way of completion. The depth, height and width of the output will be equal to
-              the input `x` divided by stride. The padding will be evenly calculated in head and tail, top and bottom,
-              left and right directions possiblily.
-              Otherwise, the last extra padding will be calculated from the tail, bottom and the right side.
+        mode (int, optional): Modes for different convolutions. Default is ``1`` . It is currently not used.
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` , ``"valid"`` or ``"pad"`` . Default: ``"valid"`` .
+
+            - ``"same"``: Pad the input around its depth/height/width dimension so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally.  If the amount is even,
+              it isuniformly distributed around the input, if it is odd, the excess amount goes
+              to the front/right/bottom side.
               If this mode is set, `pad` must be 0.
-
-            - ``"valid"``: Adopts the way of discarding. The possible largest depth, height and width of output
-              will be returned without padding. Extra pixels will be discarded. If this mode is set, `pad`
-              and `output_padding` must be 0.
-
-            - ``"pad"``: Implicit paddings on both sides of the input in depth, height and width. The number of `pad`
-              will be padded to the input Tensor borders. `pad` must be greater than or equal to 0.
-
-        pad (Union(int, tuple[int])): The pad value to be filled. Default: ``0`` . If `pad` is an integer, the paddings
-             of head, tail, top, bottom, left and right are the same, equal to pad. If `pad` is a tuple of six integers,
-             the padding of head, tail, top, bottom, left and right equal to pad[0], pad[1], pad[2], pad[3], pad[4]
-             and pad[5] correspondingly.
-        stride (Union(int, tuple[int])): The distance of kernel moving, an int number that represents
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible depth, height and width. Extra pixels that could not complete a full stride will
+              be discarded. If this mode is set, `pad` must be 0.
+            - ``"pad"``: Pad the input with a specified amount. In this mode, the amount of padding
+              in the depth, height and width dimension is determined by the `pad` parameter.
+              If this mode is set, `pad` must be greater than or equal to 0.
+
+        pad (Union(int, tuple[int]), optional): The pad value to be filled. Default: ``0`` . If `pad` is an integer,
+            the paddings of head, tail, top, bottom, left and right are the same, equal to pad.
+            If `pad` is a tuple of six integers, the padding of head, tail, top, bottom, left and right equal
+            to pad[0], pad[1], pad[2], pad[3], pad[4] and pad[5] correspondingly.
+        stride (Union(int, tuple[int]), optional): The distance of kernel moving, an int number that represents
             the depth, height and width of movement are both strides, or a tuple of three int numbers that
             represent depth, height and width of movement respectively. Default: ``1`` .
-        dilation (Union(int, tuple[int])): Specifies the space to use between kernel elements. Default: ``1`` .
-        group (int): The number of groups into which the filter is divided. `in_channels`
+        dilation (Union(int, tuple[int]), optional): Specifies the space to use between kernel elements.
+            Default: ``1`` .
+        group (int, optional): The number of groups into which the filter is divided. `in_channels`
             and `out_channels` must be divisible by `group`. Default: ``1`` .
-        output_padding (Union(int, tuple[int])): Add extra size to each dimension of the output. Default: ``0`` .
-        data_format (str): The optional value for data format. Currently only ``'NCDHW'`` is supported.
+        output_padding (Union(int, tuple[int]), optional): Add extra size to each dimension of the output.
+            Default: ``0`` .
+        data_format (str, optional): The optional value for data format. Currently only ``'NCDHW'`` is supported.
             Default: ``'NCDHW'``.
 
     Inputs:
@@ -8794,14 +8875,17 @@ class Dilation2D(Primitive):
                                       each sampling location. Its value must be greater or equal to 1 and bounded by
                                       the height and width of the input `x`.
 
-        pad_mode (str, optional): Specifies padding mode. The optional values are
-            ``"same"`` , ``"valid"`` . Default: ``"same"`` . Both upper and lower case are supported.
+        pad_mode (str, optional): Specifies the padding mode with a padding value of 0. It can be set to:
+            ``"same"`` or ``"valid"`` . Default: ``"valid"`` .
 
-            - ``"same"``: Adopts the way of completion. The height and width of the output will be the same as
-              the input `x`.
+            - ``"same"``: Pad the input around its edges so that the shape of input and output
+              are the same when `stride` is set to ``1``.
+              The amount of padding to is calculated by the operator internally, If the amount is even, it is
+              uniformly distributed around the input, if it is odd, the excess amount goes to the right/bottom side.
+            - ``"valid"``: No padding is applied to the input, and the output returns the maximum
+              possible height and width. Extra pixels that could not complete a full stride will
+              be discarded.
 
-            - ``"valid"``: Adopts the way of discarding. The possible largest height and width of output will be
-              returned without padding. Extra pixels will be discarded.
         data_format (str, optional): The value for data format, only ``'NCHW'`` is supported at present.
             Default: ``"NCHW"`` .
 
@@ -8879,7 +8963,11 @@ class Dilation2D(Primitive):
         self.pad_mode = validator.check_string(pad_mode, ['VALID', 'SAME', 'valid', 'same'], 'pad_mode', self.name)
         self.add_prim_attr('pad_mode', self.pad_mode.upper())
         self.stride = _check_format_stride_or_dilation("stride", stride, self.name, self.data_format)
-        if self.stride[2] < 1 or self.stride[2] > 255 or self.stride[3] < 1 or self.stride[3] > 255:
+
+        def is_in_range(x):
+            return 1 <= x <= 255
+
+        if not is_in_range(self.stride[2]) or not is_in_range(self.stride[3]):
             raise ValueError(f'For Dilation2D, size of stride is not supported, '
                              f'stride should be in the range of [1, 255], '
                              f'but got stride_h: `{self.stride[2]}`, stride_w: `{self.stride[3]}`.')
@@ -9418,8 +9506,8 @@ class MultilabelMarginLoss(Primitive):
             ``'sum'`` . Default: ``'mean'`` .
 
             - ``'none'``: no reduction will be applied.
-            - ``'mean'``: the sum of the output will be divided by the number of elements in the output.
-            - ``'sum'``: the output will be summed.
+            - ``'mean'``: compute and return the mean of elements in the output.
+            - ``'sum'``: the output elements will be summed.
 
     Inputs:
         - **x** (Tensor) - Predict data. Tensor of shape :math:`(C)` or :math:`(N, C)`, where :math:`N`
@@ -9428,7 +9516,7 @@ class MultilabelMarginLoss(Primitive):
           label targets padded by -1.
 
     Outputs:
-        - **y** (Union[Tensor, Scalar]) - The loss of MultilabelMarginLoss. If `reduction` is "none", its shape
+        - **y** (Union[Tensor, Scalar]) - The loss of MultilabelMarginLoss. If `reduction` is ``"none"``, its shape
           is :math:`(N)`. Otherwise, a scalar value will be returned.
         - **is_target** (Tensor) - Output tensor for backward input, with the same shape as `target`,
           data type must be int32.
@@ -9694,8 +9782,22 @@ class GridSampler3D(Primitive):
     Args:
         interpolation_mode (str, optional): An optional string specifying the interpolation method.
             The optional values are ``"bilinear"`` or ``"nearest"`` . Default: ``"bilinear"`` .
+
+            - ``"nearest"``: Nearest neighbor interpolation. Each output pixel is assigned the value of the
+              nearest input pixel. This method is simple and fast but can result in blocky or pixelated outputs.
+            - ``"bilinear"``: Bilinear interpolation. Each output pixel is a weighted average of the four nearest input
+              pixels, computed using bilinear interpolation. This method produces smoother results compared
+              to nearest neighbor interpolation.
+
         padding_mode (str, optional): An optional string specifying the pad method.
             The optional values are ``"zeros"`` , ``"border"`` or ``"reflection"`` . Default: ``"zeros"`` .
+            When the sampling grid is outside input's bounds, effects of various padding modes are as follows:
+
+            - ``"zeros"``: Pads the input tensor with zeros.
+            - ``"border"``: Pads the input tensor with the values of the pixels on the border of the tensor.
+            - ``"reflection"``: Pads the input tensor by reflecting the values of the pixels at the
+              boundary of the tensor.
+
         align_corners (bool, optional): An optional bool specifying alignment method. If set to ``True`` ,
             the extrema (-1 and 1) are considered as referring to
             the center points of the input’s corner pixels. If set to ``False`` , they are instead considered as
@@ -10178,8 +10280,12 @@ class TripletMarginLoss(Primitive):
         p (int, optional): The norm degree for pairwise distance. Default: ``2`` .
         eps (float, optional): Default: ``1e-6`` .
         swap (bool, optional): The distance swap. Default: ``False`` .
-        reduction (str, optional): Apply specific reduction method to the
-            output: ``"none"`` , ``"mean"`` , ``"sum"`` . Default: ``"mean"`` .
+        reduction (str, optional): Apply specific reduction method to the output: ``'none'`` , ``'mean'`` ,
+            ``'sum'`` . Default: ``'mean'`` .
+
+            - ``'none'``: no reduction will be applied.
+            - ``'mean'``: compute and return the mean of elements in the output.
+            - ``'sum'``: the output elements will be summed.
 
     Inputs:
         - **x** (Tensor) - A sample randomly selected from the training set. Data type must be BasicType.
@@ -10190,7 +10296,7 @@ class TripletMarginLoss(Primitive):
         - **margin** (Tensor) - Make a margin between the positive pair and the negative pair.
 
     Outputs:
-        Union[Tensor, Scalar], if `reduction` is "none", its shape is :math:`(N)`.
+        Union[Tensor, Scalar], if `reduction` is ``"none"``, its shape is :math:`(N)`.
         Otherwise, a scalar value will be returned.
 
     Raises:
@@ -10207,7 +10313,7 @@ class TripletMarginLoss(Primitive):
           is bigger than or equal to 8.
         ValueError: If length of shape of `margin` is not 0.
         ValueError: If shape of `x`, `positive` and `negative` cannot broadcast.
-        ValueError: If `reduction` is not one of 'none', 'mean', 'sum'.
+        ValueError: If `reduction` is not one of ``'none'``, ``'mean'``, ``'sum'``.
 
     Supported Platforms:
         ``GPU``
@@ -10303,6 +10409,13 @@ class GridSampler2D(Primitive):
         interpolation_mode (str, optional): An optional string specifying the interpolation method.
             The optional values are
             ``"bilinear"`` or ``"nearest"`` . Default: ``"bilinear"`` .
+
+            - ``"nearest"``: Nearest neighbor interpolation. Each output pixel is assigned the value of the
+              nearest input pixel. This method is simple and fast but can result in blocky or pixelated outputs.
+            - ``"bilinear"``: Bilinear interpolation. Each output pixel is a weighted average of the four nearest input
+              pixels, computed using bilinear interpolation. This method produces smoother results compared
+              to nearest neighbor interpolation.
+
         padding_mode (str, optional): An optional string specifying the pad method.
             The optional values are ``"zeros"`` , ``"border"`` or ``"reflection"`` . Default: ``"zeros"`` .
             When the sampling grid is outside input's bounds, effects of various padding modes are as follows:
@@ -10317,8 +10430,12 @@ class GridSampler2D(Primitive):
             and output tensors are aligned. When set to ``False`` , it is not aligned. Default: ``False`` .
 
     Inputs:
-        - **input_x** (Tensor) - A 4-D tensor with dtype of float16, float32 or float64 and shape of
-          :math:`(N, C, H_{in}, W_{in})`.
+        - **input_x** (Tensor) - A 4-D tensor with shape
+          :math:`(N, C, H_{in}, W_{in})`. Supported dtypes:
+
+          - Ascend: float16, float32.
+          - GPU/CPU: float16, float32, float64.
+
         - **grid** (Tensor) - A 4-D tensor whose dtype is the same as `input_x` and whose shape is
           :math:`(N, H_{out}, W_{out}, 2)`.
           Used to specify the sampling pixel locations normalized by the input spatial
@@ -10409,7 +10526,7 @@ class UpsampleNearest3D(Primitive):
     This operator scale up the volumetric input with specified `output_size` or `scales` factors, using nearest
     neighbor algorithm.
 
-    One of `output_size` or `scales` must be given, and can not be specified both.
+    One of `output_size` or `scales` must be given, and can not specified both at the same time.
 
     Inputs:
         - **x** (Tensor) - 5D tensor of shape :math:`(N, C, D_{in}, H_{in}, W_{in})`.
@@ -11116,46 +11233,45 @@ class Dense(Primitive):
     Applies dense connected operator for the input. The implement of the operation is as:
 
     .. math::
-        \text{output} = \text{x} * \text{w} + \text{b},
+        output = x @ w ^ T + b,
 
-    where :math:`x` is the input tensor, :math:`\text{w}` is a weight matrix with the same data type as the :math:`x` ,
-    and :math:`\text{b}` is a bias vector with the same data type as the :math:`x` (only if has_bias is True).
-
-    Args:
-        has_bias (bool): Specifies whether the layer uses a bias vector :math:`\text{b}`. Default: True.
+    where :math:`x` is the input tensor, :math:`w` is a weight matrix with the same data type as the :math:`x` ,
+    and :math:`b` is a bias vector with the same data type as the :math:`x` (only if `b` is not ``None``).
 
     Inputs:
-        - **x** (Union[Tensor, Parameter]) - The input tensor with data type of float16, float32 or float64.
-        - **w** (Union[Tensor, Parameter]) - The weight tensor with data type of float16, float32 or float64.
-        - **b** (Union[Tensor, Parameter]) - The bias tensor with data type of float16, float32 or float64.
+        - **x** (Tensor) - The shape must meet the following requirement: :math:`len(x.shape)>0`.
+        - **w** (Tensor) - The shape must meet the following requirements:
+          If :math:`len(x.shape)>1`, :math:`len(w.shape)=2`. If :math:`len(x.shape)=1`, :math:`len(w.shape)=1`.
+          :math:`w.shape[-1]=x.shape[-1]`.
+        - **b** (Union[Tensor, None]) - If `b` is not ``None``, the shape must meet the following requirements:
+          If :math:`len(x.shape)>1`, :math:`len(b.shape)=0` or :math:`len(b.shape)=1` .
+          If :math:`len(b.shape)=1`, :math:`b.shape[0]=w.shape[0]`.
+          If :math:`len(x.shape)=1`, :math:`len(b.shape)=0`.
 
     Outputs:
-        Tensor of shape :math:`(*x.shape[:-1], w.shape[0])`.
-
-    Raises:
-        TypeError: If `has_bias` is not a bool.
+        If :math:`len(x.shape)>1`, Tensor of shape :math:`(*x.shape[:-1], w.shape[0])`.
+        If :math:`len(x.shape)=1`, Tensor of shape :math:`()`.
 
     Supported Platforms:
-        ``GPU``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
-        >>> from mindspore.ops.operations import nn_ops
+        >>> import numpy as np
+        >>> from mindspore import Tensor, ops
         >>> x = Tensor(np.random.random((4, 5, 6, 7)).astype(np.float32))
-        >>> weight = Parameter(np.random.random((6, 7)).astype(np.float32))
-        >>> bias = Parameter(np.random.random((6,)).astype(np.float32))
-        >>> dense = nn_ops.Dense()
+        >>> weight = Tensor(np.random.random((6, 7)).astype(np.float32))
+        >>> bias = Tensor(np.random.random((6,)).astype(np.float32))
+        >>> dense = ops.Dense()
         >>> output = dense(x, weight, bias)
         >>> print(output.shape)
         (4, 5, 6, 6)
     """
 
     @prim_attr_register
-    def __init__(self, has_bias=True):
+    def __init__(self):
         """Initialize Dense."""
         self.init_prim_io_names(inputs=['x', 'w', 'b'], outputs=["output"])
-        self.has_bias = has_bias
-        self.has_bias = validator.check_bool(has_bias, "has_bias", "Dense")
-        self.add_prim_attr("has_bias", self.has_bias)
+        self.add_prim_attr("has_bias", True)
 
 
 class WKV(Primitive):
@@ -11166,22 +11282,22 @@ class WKV(Primitive):
 
     Inputs:
         - **w** (Tensor) - The time_first tensor with data type of float32.
-          Input tensor of shape :math:`(hidden_size,)`.
+          Input tensor of shape :math:`(hidden\_size,)`.
         - **u** (Tensor]) - The time_decay tensor with data type of float32.
-          Input tensor of shape :math:`(hidden_size,)`.
+          Input tensor of shape :math:`(hidden\_size,)`.
         - **k** (Tensor) - The key tensor with data type of float32.
-          Input tensor of shape :math:`(batch_size, seq_length, hidden_size)`.
+          Input tensor of shape :math:`(batch\_size, seq\_length, hidden\_size)`.
         - **v** (Tensor) - The value tensor with data type of float32.
-          Input tensor of shape :math:`(batch_size, seq_length, hidden_size)`.
+          Input tensor of shape :math:`(batch\_size, seq\_length, hidden\_size)`.
         - **sp** (Tensor) - The states_p tensor with data type of float32.
-          Input tensor of shape :math:`(batch_size, seq_length, hidden_size)`.
+          Input tensor of shape :math:`(batch\_size, seq\_length, hidden\_size)`.
         - **sq** (Tensor) - The states_q tensor with data type of float32.
-          Input tensor of shape :math:`(batch_size, hidden_size)`.
+          Input tensor of shape :math:`(batch\_size, hidden\_size)`.
         - **sm** (Tensor) - The states_m tensor with data type of float32.
-          Input tensor of shape :math:`(batch_size, hidden_size)`.
+          Input tensor of shape :math:`(batch\_size, hidden\_size)`.
 
     Outputs:
-        Tensor of shape :math:`(batch_size, seq_length, hidden_size)`.
+        Tensor of shape :math:`(batch\_size, seq\_length, hidden\_size)`.
 
     Supported Platforms:
         ``Ascend``
@@ -11209,3 +11325,255 @@ class WKV(Primitive):
         """Initialize WKV."""
         self.init_prim_io_names(inputs=["time_first", "time_decay", "key", "value", "sp", "sq", "sm"],
                                 outputs=["output", "out_sp", "out_sq", "out_sm"])
+
+
+class PromptFlashAttention(Primitive):
+    r"""
+    The interface for fully inference.
+    B -- Batch size
+    S -- Sequence length
+    H -- Hidden size
+
+    Refer to :func:mindspore.ops.prompt_flash_attention for more detail.
+
+    .. warning::
+        This is an experimental API that is subject to change or deletion.
+
+    Args:
+        num_heads (int): The number of heads.
+        scale_value (float): The scale value indicating the scale coefficient, which is used as the scalar of
+          Muls in the calculation. Default: 1.0.
+        pre_tokens (int): Previous tokens. Default: 2147483547.
+        next_tokens (int): next tokens.  Default: 0.
+          indicate the upper triangle, Indicate the number of data blocks involved in the calculation. The value 0
+          indicates that the data blocks in the upper triangle are not involved in the calculation
+        input_layout (str): the data layout of the input qkv, support `(BSH)` and `(BNSD)`, Default `BSH`.
+        num_key_value_heads (int): head numbers of key/value which are used in GQA algorithm.
+          The value o indicates if the key and value have the same head nums, use numHeads.  Default: 0.
+        sparse_mode (int): Default: 0
+
+    Inputs:
+        - **query** (Tensor) - The query tensor with data type of float16 or float32.
+          Input tensor of shape :math:`(B, S, H)` / `(B, N, S, D)`.
+        - **key** (Tensor) - The key tensor with data type of float16 or float32.
+          Input tensor of shape :math:`(B, S, H)` / `(B, N, S, D)`.
+        - **value** (Tensor) - The value tensor with data type of float16 or float32.
+          Input tensor of shape :math:`(B, S, H)` / `(B, N, S, D)`.
+        - **attn_mask** (Tensor) - The attention mask tensor with data type of float16 or float32.
+          For each element, 0 indicates retention and 1 indicates discard. Input tensor of shape :math:`(B, 1, S, S)`.
+        - **actual_seq_lengths** (Tensor): Describe actual sequence length of each input with data type of int.
+        - **actual_seq_lengths_kv** (Tensor): Describe actual sequence length of each input with data type of int.
+        - **padding_mask** (Tensor) - The padding mask tensor with data type of float16 or float32
+        - **dep_scale1** (Tensor)
+        - **quant_scale1** (Tensor)
+        - **deq_scale2** (Tensor)
+        - **quant_scale2** (Tensor)
+        - **quant_offset2** (Tensor)
+
+
+    Outputs:
+        - **attention_out** (Tensor) - Input tensor of shape :math:`(B, S, H)` / `(B, N, S, D)`.
+
+        Supported Platforms:
+        ``Ascend``
+
+    Examples:
+        >>> import mindspore.ops.operations.nn_ops as P
+        >>> from mindspore import Tensor
+        >>> import numpy as np
+        >>> B = 1
+        >>> N = 16
+        >>> S = 256
+        >>> D = 16
+        >>> query = Tensor(np.ones((B, N, S, D), dtype=np.float16))
+        >>> key = Tensor(np.ones((B, N, S, D), dtype=np.float16))
+        >>> value = Tensor(np.ones((B, N, S, D), dtype=np.float16))
+        >>> pfa = P.PromptFlashAttention(N, input_layout='BNSD')
+        >>> out = pfa(query, key, value, None, None, None, None, None, None, None, None, None)
+        >>> print(out[0].shape)
+        (1, 16, 256, 16)
+    """
+
+    @prim_attr_register
+    def __init__(self, num_heads, scale_value=1.0, pre_tokens=2147483547, next_tokens=0, input_layout='BSH',
+                 num_key_value_heads=0, sparse_mode=0):
+        """Initialize PromptFlashAttention."""
+        validator.check_value_type('num_heads', num_heads, [int], self.name)
+        validator.check_value_type('scale_value', scale_value, [float], self.name)
+        validator.check_value_type('pre_tokens', pre_tokens, [int], self.name)
+        validator.check_value_type('next_tokens', next_tokens, [int], self.name)
+        validator.check_value_type('input_layout', input_layout, [str], self.name)
+        validator.check_value_type('num_key_value_heads', num_key_value_heads, [int], self.name)
+        validator.check_value_type('sparse_mode', sparse_mode, [int], self.name)
+        self.init_prim_io_names(inputs=["query", "key", "value", "attn_mask", "actual_seq_lengths",
+                                        "actual_seq_lengths_kv", "padding_mask", "deq_scale1", "quant_scale1",
+                                        "deq_scale2", "quant_scale2", "quant_offset2"],
+                                outputs=["attention_out"])
+
+
+class FlashAttentionScore(Primitive):
+    r"""
+    FlashAttentionScore.
+    .. warning::
+        This is an experimental API that is subject to change or deletion.
+    B -- Batch size
+    S1 -- Sequence length of query
+    S2 -- Sequence length of key and value
+    N1 -- Num heads of query
+    N2 -- Num heads of key and value, and N2 must be a factor of N1
+    D -- head size
+    H1 -- Hidden size of query, which equals to N1 * D
+    H2 -- Hidden size of key and value, which equals to N2 * D
+    Args:
+        head_num (int): The head num of query.
+        keep_prob (float): The keep probability of dropout. Default: 1.0.
+        scale_value (float): The scale value. Default: 1.0.
+        pre_tokens (int): Previous tokens. Default: 65536.
+        next_tokens (int): Next tokens. Default: 65536.
+        inner_precise (int): Specify the execution mode, where 0 indicates high precision mode and 1 indicates high
+        performance mode. Only support 0 currently. Default: 0.
+        input_layout (str, optional): Specifies the layout of `query`, the value must be one of ["BSH", "BNSD"].
+        Default: "BSH".
+        sparse_mode (int): Default 0.
+
+    Inputs:
+        - **query** (Tensor[float16, float32, bfloat16]) - The query tensor.
+          Input tensor of shape :math:`(B, S1, H1)` or `(B, N1, S1, D)`.
+        - **key** (Tensor[float16, float32, bfloat16]) - The key tensor.
+          Input tensor of shape :math:`(B, S2, H2)` or `(B, N2, S2, D)`.
+        - **value** (Tensor[float16, float32, bfloat16]) - The value tensor.
+          Input tensor of shape :math:`(B, S2, H2)` or `(B, N2, S2, D)`.
+        - **real_shift** (Tensor[float16, float32, bfloat16], None) - The position embedding code.
+          Input tensor of shape :math: `(B, N1, S1, S2)` or `(B, N1, 1, S2)`.
+        - **drop_mask** (Tensor[uint8], None) - The dropout mask tensor.
+          Input tensor of shape :math:`(B, N1, S1, S2 // 8) or None`.
+        - **padding_mask** (None) - The padding mask of float16 or float32, not implemented yet.
+        - **attn_mask** (Tensor[uint8], None) - The attention mask tensor.
+          For each element, 0 indicates retention and 1 indicates discard.
+          Input tensor of shape :math:`(B, N1, S1, S2)`, `(B, 1, S1, S2)` or `(S1, S2)`.
+        - **prefix** (Tensor[int64], None) - Not implemented yet.
+          Input tensor of shape :math:`(B,)`.
+
+    Outputs:
+        - **softmax_max** (Tensor[float32]) - (B, N1, S1, 8)
+        - **softmax_sum** (Tensor[float32]) - (B, N1, S1, 8)
+        - **softmax_out** (Tensor[float32]) - Useless output, ignore it. Output tensor of shape : `()`
+        - **attention_out** (Tensor[float16, float32, bfloat16]) - The output of attention, its shape, and data type
+          are the same as the query.
+
+    Supported Platforms:
+        ``Ascend``
+    """
+
+    @prim_attr_register
+    def __init__(self, head_num, keep_prob=1.0, scale_value=1.0, pre_tokens=65536, next_tokens=65536, inner_precise=0,
+                 input_layout="BSH", sparse_mode=0):
+        """Initialize FlashAttentionScore"""
+        validator.check_value_type('head_num', head_num, [int], self.name)
+        validator.check_value_type('keep_prob', keep_prob, [int, float], self.name)
+        validator.check_float(keep_prob, 0.0, validator.GE, "keep_prob", self.name)
+        validator.check_float(keep_prob, 1.0, validator.LE, "keep_prob", self.name)
+        validator.check_value_type('scale_value', scale_value, [float], self.name)
+        validator.check_value_type('pre_tokens', pre_tokens, [int], self.name)
+        validator.check_value_type('next_tokens', next_tokens, [int], self.name)
+        validator.check_value_type('inner_precise', inner_precise, [int], self.name)
+        validator.check_value_type('sparse_mode', sparse_mode, [int], self.name)
+        if inner_precise not in [0]:
+            raise ValueError(f"Attribute 'inner_precise' must be 0, but got {inner_precise}")
+        validator.check_value_type('input_layout', input_layout, [str], self.name)
+        if input_layout not in ["BSH", "BNSD"]:
+            raise ValueError(f"Attribute 'input_layout' must be either 'BSH' or 'BNSD', but got {input_layout}")
+        self.init_prim_io_names(
+            inputs=['query', 'key', 'value', 'real_shift', 'drop_mask', 'padding_mask', 'attn_mask', 'prefix'],
+            outputs=['softmax_max', 'softmax_sum', 'softmax_out', 'attention_out'])
+
+
+class RmsNorm(Primitive):
+    r"""
+    The RmsNorm operator is a normalization operation, and its formula is:
+
+    .. math::
+        y=\frac{x_i}{\sqrt{\frac{1}{n}}\sum_{i=1}^{n}{ x_i^2}+\varepsilon  }\gamma_i
+
+    .. warning::
+        This is an experimental API that is subject to change or deletion.
+
+    Args:
+        epsilon (float): prevent division by 0, default value is `1e-6`
+
+    Inputs:
+        - **input_x** (Tensor) - Input data of RmsNorm, support data type: float16, float32, bfloat16.
+        - **gamma** (Tensor) - Support data type: float16, float32, bfloat16.
+
+    Outputs:
+        - **y** (Tensor) - Has the same type and shape with `input_x`.
+        - **rstd** (Tensor) - Has the same type with `input_x`, used by gradient calculation.
+
+    Raises:
+        TypeError: If data type of `input_x` is not one of the following: float16, float32, bfloat16.
+        TypeError: If data type of `gamma` is not one of the following: float16, float32, bfloat16.
+        TypeError: If data type of "input_x" is not the same with the data type of "gamma"
+
+    Supported Platforms:
+        ``Ascend``
+    """
+
+    @prim_attr_register
+    def __init__(self, epsilon=1e-6):
+        """Initialize Dense."""
+        validator.check_value_type("epsilon", epsilon, [float], self.name)
+        self.init_prim_io_names(inputs=['x', 'gamma'], outputs=["y", "rstd"])
+
+
+class PagedAttention(Primitive):
+    r"""
+    .. warning::
+        This is an experimental API that is subject to change or deletion.
+    """
+    @prim_attr_register
+    def __init__(self, head_num, scale_value=1.0, kv_head_num=0):
+        """Initialize PagedAttention"""
+        validator.check_value_type('head_num', head_num, [int], self.name)
+        validator.check_value_type('scale_value', scale_value, [float], self.name) # scale after qkbmm
+        validator.check_value_type('kv_head_num', kv_head_num, [int], self.name) # for MQA
+        self.init_prim_io_names(
+            inputs=['query', 'key_cache', 'value_cache', 'block_tables', 'context_lens'],
+            outputs=['attention_out'])
+
+
+class PagedAttentionMask(Primitive):
+    r"""
+    .. warning::
+        This is an experimental API that is subject to change or deletion.
+    """
+    @prim_attr_register
+    def __init__(self, head_num, scale_value=1.0, kv_head_num=0):
+        """Initialize PagedAttentionMask"""
+        validator.check_value_type('head_num', head_num, [int], self.name)
+        validator.check_value_type('scale_value', scale_value, [float], self.name) # scale after qkbmm
+        validator.check_value_type('kv_head_num', kv_head_num, [int], self.name) # for MQA
+        self.init_prim_io_names(
+            inputs=['query', 'key_cache', 'value_cache', 'block_tables', 'context_lens', 'alibi_mask'],
+            outputs=['attention_out'])
+
+
+class ReshapeAndCache(Primitive):
+    r"""
+    .. warning::
+        This is an experimental API that is subject to change or deletion.
+    """
+    __mindspore_signature__ = (
+        sig.make_sig('key', dtype=sig.sig_dtype.T),
+        sig.make_sig('value', dtype=sig.sig_dtype.T),
+        sig.make_sig('key_cache', sig.sig_rw.RW_WRITE, dtype=sig.sig_dtype.T),
+        sig.make_sig('value_cache', sig.sig_rw.RW_WRITE, dtype=sig.sig_dtype.T),
+        sig.make_sig('slot_mapping', dtype=sig.sig_dtype.T1),
+    )
+
+    @prim_attr_register
+    def __init__(self):
+        """Initialize ReshapeAndCache"""
+        self.init_prim_io_names(
+            inputs=['key', 'value', 'key_cache', 'value_cache', 'slot_mapping'],
+            outputs=['key_out'])
+        self.add_prim_attr('side_effect_mem', True)