docs: add missing docstrings, Raises sections, and fix Args in networ…

monai/networks/blocks/crossattention.py

@@ -63,6 +63,13 @@ def __init__(
             attention_dtype: cast attention operations to this dtype.
             use_flash_attention: if True, use Pytorch's inbuilt flash attention for a memory efficient attention mechanism
                 (see https://pytorch.org/docs/2.2/generated/torch.nn.functional.scaled_dot_product_attention.html).
+
+        Raises:
+            ValueError: if ``dropout_rate`` is not between 0 and 1.
+            ValueError: if ``hidden_size`` is not divisible by ``num_heads`` when ``dim_head`` is not set.
+            ValueError: if ``causal`` is True and ``sequence_length`` is not provided.
+            ValueError: if both ``save_attn`` and ``use_flash_attention`` are True.
+            ValueError: if ``rel_pos_embedding`` is not None and ``use_flash_attention`` is True.
         """
 
         super().__init__()

monai/networks/blocks/downsample.py

@@ -127,6 +127,11 @@ def __init__(
                 Only used in the "maxpool", "avgpool" or "pixelunshuffle" modes.
             post_conv: a conv block applied after downsampling. Defaults to None. Only used in the "maxpool" and "avgpool" modes.
             bias: whether to have a bias term in the default preconv and conv layers. Defaults to True.
+
+        Raises:
+            ValueError: if ``mode`` is ``"conv"`` or ``"convgroup"`` and ``in_channels`` is not specified.
+            ValueError: if ``mode`` is ``"maxpool"`` or ``"avgpool"``, ``pre_conv`` is ``"default"``,
+                ``out_channels != in_channels``, and ``in_channels`` is not specified.
         """
         super().__init__()
 
@@ -261,6 +266,10 @@ def __init__(
                 When ``conv_block`` is an ``nn.module``,
                 please ensure the input number of channels matches requirements.
             bias: whether to have a bias term in the default conv_block. Defaults to True.
+
+        Raises:
+            ValueError: if ``scale_factor`` is not greater than 0.
+            ValueError: if ``conv_block`` is ``"default"`` and ``in_channels`` is not specified.
         """
         super().__init__()
 

monai/networks/blocks/localnet_block.py

@@ -72,10 +72,23 @@ def get_deconv_block(spatial_dims: int, in_channels: int, out_channels: int) ->
 
 
 class ResidualBlock(nn.Module):
+    """
+    A residual block with two convolutional layers and a skip connection.
+    """
 
     def __init__(
         self, spatial_dims: int, in_channels: int, out_channels: int, kernel_size: Sequence[int] | int
     ) -> None:
+        """
+        Args:
+            spatial_dims: number of spatial dimensions.
+            in_channels: number of input channels. Must equal ``out_channels``.
+            out_channels: number of output channels. Must equal ``in_channels``.
+            kernel_size: convolution kernel size.
+
+        Raises:
+            ValueError: if ``in_channels != out_channels``.
+        """
         super().__init__()
         if in_channels != out_channels:
             raise ValueError(
@@ -96,8 +109,20 @@ def forward(self, x) -> torch.Tensor:
 
 
 class LocalNetResidualBlock(nn.Module):
+    """
+    A residual block used in LocalNet that adds a mid-level feature to the convolution output.
+    """
 
     def __init__(self, spatial_dims: int, in_channels: int, out_channels: int) -> None:
+        """
+        Args:
+            spatial_dims: number of spatial dimensions.
+            in_channels: number of input channels. Must equal ``out_channels``.
+            out_channels: number of output channels. Must equal ``in_channels``.
+
+        Raises:
+            ValueError: if ``in_channels != out_channels``.
+        """
         super().__init__()
         if in_channels != out_channels:
             raise ValueError(
@@ -270,11 +295,15 @@ def __init__(
     ) -> None:
         """
         Args:
-        spatial_dims: number of spatial dimensions.
-        in_channels: number of input channels.
-        out_channels: number of output channels.
-        act: activation type and arguments. Defaults to ReLU.
-        kernel_initializer: kernel initializer. Defaults to None.
+            spatial_dims: number of spatial dimensions.
+            in_channels: number of input channels.
+            out_channels: number of output channels.
+            act: activation type and arguments. Defaults to ReLU.
+            initializer: kernel initializer, either ``"kaiming_uniform"`` or ``"zeros"``. Defaults to
+                ``"kaiming_uniform"``. Note: ``"kaiming_uniform"`` uses ``nn.init.kaiming_normal_`` internally.
+
+        Raises:
+            ValueError: if ``initializer`` is not ``"kaiming_uniform"`` or ``"zeros"``.
         """
         super().__init__()
         self.conv_block = get_conv_block(

monai/networks/blocks/patchembedding.py

@@ -68,6 +68,12 @@ def __init__(
             spatial_dims: number of spatial dimensions.
             pos_embed_kwargs: additional arguments for position embedding. For `sincos`, it can contain
                               `temperature` and for fourier it can contain `scales`.
+
+        Raises:
+            ValueError: if ``dropout_rate`` is not between 0 and 1.
+            ValueError: if ``hidden_size`` is not divisible by ``num_heads``.
+            ValueError: if any dimension of ``patch_size`` is larger than the corresponding ``img_size`` dimension.
+            ValueError: if ``proj_type`` is ``"perceptron"`` and ``patch_size`` does not evenly divide ``img_size``.
         """
 
         super().__init__()

monai/networks/blocks/selfattention.py

@@ -66,6 +66,13 @@ def __init__(
             use_flash_attention: if True, use Pytorch's inbuilt flash attention for a memory efficient attention mechanism
                 (see https://pytorch.org/docs/2.2/generated/torch.nn.functional.scaled_dot_product_attention.html).
 
+        Raises:
+            ValueError: if ``dropout_rate`` is not between 0 and 1.
+            ValueError: if ``hidden_size`` is not divisible by ``num_heads``.
+            ValueError: if ``causal`` is True and ``sequence_length`` is not provided.
+            ValueError: if both ``save_attn`` and ``use_flash_attention`` are True.
+            ValueError: if ``rel_pos_embedding`` is not None and ``use_flash_attention`` is True.
+
         """
 
         super().__init__()

monai/networks/blocks/upsample.py

@@ -86,6 +86,11 @@ def __init__(
                 size of `scale_factor` with a stride of 1. See also: :py:class:`monai.networks.blocks.SubpixelUpsample`.
                 Only used in the "pixelshuffle" mode.
 
+        Raises:
+            ValueError: if ``mode`` is ``"deconv"`` or ``"deconvgroup"`` and ``in_channels`` is not specified.
+            ValueError: if ``mode`` is ``"nontrainable"``, ``pre_conv`` is not set, and
+                ``out_channels != in_channels``.
+
         """
         super().__init__()
         scale_factor_ = ensure_tuple_rep(scale_factor, spatial_dims)