diff --git a/mindcv/models/__init__.py b/mindcv/models/__init__.py
index f3395796c..fd6b47d9d 100644
--- a/mindcv/models/__init__.py
+++ b/mindcv/models/__init__.py
@@ -52,6 +52,7 @@
     vgg,
     visformer,
     vit,
+    vit_sam,
     volo,
     xception,
     xcit,
@@ -109,6 +110,7 @@
 from .vgg import *
 from .visformer import *
 from .vit import *
+from .vit_sam import *
 from .volo import *
 from .xception import *
 from .xcit import *
@@ -168,6 +170,7 @@
 __all__.extend(vgg.__all__)
 __all__.extend(visformer.__all__)
 __all__.extend(vit.__all__)
+__all__.extend(vit_sam.__all__)
 __all__.extend(volo.__all__)
 __all__.extend(["Xception", "xception"])
 __all__.extend(xcit.__all__)
diff --git a/mindcv/models/layers/format.py b/mindcv/models/layers/format.py
index 058a74517..167cca497 100644
--- a/mindcv/models/layers/format.py
+++ b/mindcv/models/layers/format.py
@@ -1,5 +1,4 @@
 from enum import Enum
-from typing import Union
 
 import mindspore
 
@@ -11,9 +10,6 @@ class Format(str, Enum):
     NLC = 'NLC'
 
 
-FormatT = Union[str, Format]
-
-
 def nchw_to(x: mindspore.Tensor, fmt: Format):
     if fmt == Format.NHWC:
         x = x.permute(0, 2, 3, 1)
diff --git a/mindcv/models/layers/patch_dropout.py b/mindcv/models/layers/patch_dropout.py
index ad854dbfc..03bf65b47 100644
--- a/mindcv/models/layers/patch_dropout.py
+++ b/mindcv/models/layers/patch_dropout.py
@@ -8,41 +8,41 @@ class PatchDropout(nn.Cell):
     """
     https://arxiv.org/abs/2212.00794
     """
+
     def __init__(
-            self,
-            prob: float = 0.5,
-            num_prefix_tokens: int = 1,
-            ordered: bool = False,
-            return_indices: bool = False,
+        self,
+        prob: float = 0.5,
+        num_prefix_tokens: int = 1,
+        ordered: bool = False,
+        return_indices: bool = False,
     ):
         super().__init__()
-        assert 0 <= prob < 1.
+        assert 0 <= prob < 1.0
         self.prob = prob
         self.num_prefix_tokens = num_prefix_tokens  # exclude CLS token (or other prefix tokens)
         self.ordered = ordered
         self.return_indices = return_indices
-        self.sort = ops.Sort()
 
-    def forward(self, x):
-        if not self.training or self.prob == 0.:
+    def construct(self, x):
+        if not self.training or self.prob == 0.0:
             if self.return_indices:
                 return x, None
             return x
 
         if self.num_prefix_tokens:
-            prefix_tokens, x = x[:, :self.num_prefix_tokens], x[:, self.num_prefix_tokens:]
+            prefix_tokens, x = x[:, : self.num_prefix_tokens], x[:, self.num_prefix_tokens :]
         else:
             prefix_tokens = None
 
         B = x.shape[0]
         L = x.shape[1]
-        num_keep = max(1, int(L * (1. - self.prob)))
-        _, indices = self.sort(ms.Tensor(np.random.rand(B, L)).astype(ms.float32))
+        num_keep = max(1, int(L * (1.0 - self.prob)))
+        _, indices = ops.sort(ms.Tensor(np.random.rand(B, L)).astype(ms.float32))
         keep_indices = indices[:, :num_keep]
         if self.ordered:
             # NOTE does not need to maintain patch order in typical transformer use,
             # but possibly useful for debug / visualization
-            keep_indices, _ = self.sort(keep_indices)
+            keep_indices, _ = ops.sort(keep_indices)
         keep_indices = ops.broadcast_to(ops.expand_dims(keep_indices, axis=-1), (-1, -1, x.shape[2]))
         x = ops.gather_elements(x, dim=1, index=keep_indices)
 
diff --git a/mindcv/models/layers/patch_embed.py b/mindcv/models/layers/patch_embed.py
index 661e07890..315f1fb89 100644
--- a/mindcv/models/layers/patch_embed.py
+++ b/mindcv/models/layers/patch_embed.py
@@ -18,6 +18,7 @@ class PatchEmbed(nn.Cell):
         embed_dim (int): Number of linear projection output channels. Default: 96.
         norm_layer (nn.Cell, optional): Normalization layer. Default: None
     """
+
     output_fmt: Format
 
     def __init__(
@@ -37,11 +38,11 @@ def __init__(
         self.patch_size = to_2tuple(patch_size)
         if image_size is not None:
             self.image_size = to_2tuple(image_size)
-            self.patches_resolution = tuple([s // p for s, p in zip(self.image_size, self.patch_size)])
-            self.num_patches = self.patches_resolution[0] * self.patches_resolution[1]
+            self.grid_size = tuple([s // p for s, p in zip(self.image_size, self.patch_size)])
+            self.num_patches = self.grid_size[0] * self.grid_size[1]
         else:
             self.image_size = None
-            self.patches_resolution = None
+            self.grid_size = None
             self.num_patches = None
 
         if output_fmt is not None:
@@ -86,8 +87,8 @@ def construct(self, x: Tensor) -> Tensor:
         # FIXME look at relaxing size constraints
         x = self.proj(x)
         if self.flatten:
-            x = ops.Reshape()(x, (B, self.embed_dim, -1))  # B Ph*Pw C
-            x = ops.Transpose()(x, (0, 2, 1))
+            x = ops.reshape(x, (B, self.embed_dim, -1))  # B Ph*Pw C
+            x = ops.transpose(x, (0, 2, 1))
         elif self.output_fmt != "NCHW":
             x = nchw_to(x, self.output_fmt)
         if self.norm is not None:
diff --git a/mindcv/models/layers/pos_embed.py b/mindcv/models/layers/pos_embed.py
index ba4548580..8c646d40b 100644
--- a/mindcv/models/layers/pos_embed.py
+++ b/mindcv/models/layers/pos_embed.py
@@ -11,11 +11,11 @@
 
 
 def resample_abs_pos_embed(
-        posemb,
-        new_size: List[int],
-        old_size: Optional[List[int]] = None,
-        num_prefix_tokens: int = 1,
-        interpolation: str = 'nearest',
+    posemb,
+    new_size: List[int],
+    old_size: Optional[List[int]] = None,
+    num_prefix_tokens: int = 1,
+    interpolation: str = 'nearest',
 ):
     # sort out sizes, assume square if old size not provided
     num_pos_tokens = posemb.shape[1]
diff --git a/mindcv/models/vit_sam.py b/mindcv/models/vit_sam.py
new file mode 100644
index 000000000..b5c21eb74
--- /dev/null
+++ b/mindcv/models/vit_sam.py
@@ -0,0 +1,594 @@
+from collections import OrderedDict
+from functools import partial
+from typing import Callable, Optional, Tuple
+
+import numpy as np
+
+import mindspore as ms
+from mindspore import Parameter, nn, ops
+from mindspore.common.initializer import initializer
+
+from .helpers import load_pretrained
+from .layers.compatibility import Dropout
+from .layers.drop_path import DropPath
+from .layers.format import Format
+from .layers.mlp import Mlp
+from .layers.patch_dropout import PatchDropout
+from .layers.patch_embed import PatchEmbed
+from .layers.pooling import GlobalAvgPooling
+from .registry import register_model
+
+__all__ = [
+    "VitSAM",
+    "samvit_base_patch16",
+    "samvit_large_patch16",
+    "samvit_huge_patch16",
+]
+
+
+def _cfg(url="", **kwargs):
+    return {
+        "url": url,
+        "num_classes": 1000,
+        "input_size": (3, 1024, 1024),
+        "first_conv": "patch_embed.proj",
+        "classifier": "head.classifier",
+        **kwargs,
+    }
+
+
+default_cfgs = {
+    "samvit_base_patch16": _cfg(
+        url="", num_classes=0, input_size=(3, 1024, 1024), crop_pct=1.0
+    ),
+    "samvit_large_patch16": _cfg(
+        url="", num_classes=0, input_size=(3, 1024, 1024), crop_pct=1.0
+    ),
+    "samvit_huge_patch16": _cfg(
+        url="", num_classes=0, input_size=(3, 1024, 1024), crop_pct=1.0
+    ),
+}
+
+
+class Attention(nn.Cell):
+    """Multi-head Attention block with relative position embeddings."""
+
+    def __init__(
+        self,
+        dim,
+        num_heads=8,
+        qkv_bias=True,
+        qk_norm=False,
+        attn_drop=0.,
+        norm_layer=nn.LayerNorm,
+        use_rel_pos: bool = False,
+        input_size: Optional[Tuple[int, int]] = None,
+    ):
+        super().__init__()
+        assert dim % num_heads == 0, "dim should be divisible by num_heads"
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+        self.scale = self.head_dim ** -0.5
+
+        self.qkv = nn.Dense(dim, dim * 3, has_bias=qkv_bias)
+        self.q_norm = norm_layer((self.head_dim,)) if qk_norm else nn.Identity()
+        self.k_norm = norm_layer((self.head_dim,)) if qk_norm else nn.Identity()
+        self.attn_drop = Dropout(attn_drop)
+        self.proj = nn.Dense(dim, dim)
+
+        self.use_rel_pos = use_rel_pos
+        if self.use_rel_pos:
+            assert (
+                    input_size is not None
+            ), "Input size must be provided if using relative positional encoding."
+            # initialize relative positional embeddings
+            self.rel_pos_h = ms.Parameter(ops.zeros((2 * input_size[0] - 1, self.head_dim)))
+            self.rel_pos_w = ms.Parameter(ops.zeros((2 * input_size[1] - 1, self.head_dim)))
+
+        self.attn_matmul_v = ops.BatchMatMul()
+        self.q_matmul_k = ops.BatchMatMul(transpose_b=True)
+
+    def construct(self, x):
+        b, h, w, _ = x.shape
+        x = ops.reshape(x, (b, h * w, -1))
+
+        qkv = ops.reshape(self.qkv(x), (b, h * w, 3, self.num_heads, -1))
+        qkv = ops.transpose(qkv, (2, 0, 3, 1, 4))
+
+        qkv = ops.reshape(qkv, (3, b * self.num_heads, h * w, -1))
+        q, k, v = ops.unstack(qkv, axis=0)
+        q, k = self.q_norm(q), self.k_norm(k)
+
+        if self.use_rel_pos:
+            attn_bias = get_decomposed_rel_pos(q, self.rel_pos_h, self.rel_pos_w, (h, w), (h, w))
+        else:
+            attn_bias = None
+
+        q = ops.mul(q, self.scale)
+        attn = self.q_matmul_k(q, k)
+
+        if attn_bias is not None:
+            attn = attn + attn_bias
+        attn = ops.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+
+        out = self.attn_matmul_v(attn, v)
+        out = ops.reshape(out, (b, self.num_heads, h * w, -1))
+        out = ops.transpose(out, (0, 2, 1, 3))
+        out = ops.reshape(out, (b, h * w, -1))
+        out = self.proj(out)
+        out = ops.reshape(out, (b, h, w, -1))
+        return out
+
+
+def get_rel_pos(q_size: int, k_size: int, rel_pos: ms.Tensor) -> ms.Tensor:
+    """
+    Get relative positional embeddings according to the relative positions of
+        query and key sizes.
+    Args:
+        q_size (int): size of query q.
+        k_size (int): size of key k.
+        rel_pos (Tensor): relative position embeddings (L, C).
+
+    Returns:
+        Extracted positional embeddings according to relative positions.
+    """
+    max_rel_dist = int(2 * max(q_size, k_size) - 1)
+    # Interpolate rel pos if needed.
+    if rel_pos.shape[0] != max_rel_dist:
+        # Interpolate rel pos.
+        rel_pos_resized = ops.interpolate(
+            rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1),
+            size=max_rel_dist,
+            mode="linear",
+        )
+        rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).permute(1, 0)
+    else:
+        rel_pos_resized = rel_pos
+
+    # Scale the coords with short length if shapes for q and k are different.
+    q_coords = ops.arange(q_size)[:, None] * max(k_size / q_size, 1.0)
+    k_coords = ops.arange(k_size)[None, :] * max(q_size / k_size, 1.0)
+    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)
+
+    return rel_pos_resized[relative_coords.long()]
+
+
+def get_decomposed_rel_pos(
+    q: ms.Tensor,
+    rel_pos_h: ms.Tensor,
+    rel_pos_w: ms.Tensor,
+    q_size: Tuple[int, int],
+    k_size: Tuple[int, int],
+) -> ms.Tensor:
+    """
+    Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
+    https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py   # noqa B950
+    Args:
+        attn (Tensor): attention map.
+        q (Tensor): query q in the attention layer with shape (B, q_h * q_w, C).
+        rel_pos_h (Tensor): relative position embeddings (Lh, C) for height axis.
+        rel_pos_w (Tensor): relative position embeddings (Lw, C) for width axis.
+        q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
+        k_size (Tuple): spatial sequence size of key k with (k_h, k_w).
+
+    Returns:
+        attn (Tensor): attention map with added relative positional embeddings.
+    """
+    q_h, q_w = q_size
+    k_h, k_w = k_size
+    rh = get_rel_pos(q_h, k_h, rel_pos_h)
+    rw = get_rel_pos(q_w, k_w, rel_pos_w)
+
+    b, _, dim = q.shape
+    r_q = q.reshape(b, q_h, q_w, dim)
+    dtype = r_q.dtype
+    # rel_h = ops.einsum("bhwc,hkc->bhwk", r_q, Rh)
+    rel_h = ops.BatchMatMul(transpose_b=True)(r_q, ops.unsqueeze(rh, 0).astype(dtype).repeat(b, axis=0))
+    # rel_w = ops.einsum("bhwc,wkc->bhwk", r_q, Rw)
+    rel_w = ops.mul(ops.unsqueeze(r_q, -2), ops.unsqueeze(ops.unsqueeze(rw, 0), 0).astype(dtype)).sum(axis=-1)
+
+    attn_bias = rel_h[:, :, :, :, None] + rel_w[:, :, :, None, :]
+    attn_bias = ops.reshape(attn_bias, (-1, q_h * q_w, k_h * k_w))
+    return attn_bias
+
+
+class Block(nn.Cell):
+    """Transformer blocks with support of window attention and residual propagation blocks"""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int,
+        mlp_ratio: float = 4.0,
+        qkv_bias: bool = True,
+        qk_norm: bool = False,
+        proj_drop: float = 0.,
+        attn_drop: float = 0.,
+        init_values: Optional[float] = None,
+        drop_path: float = 0.,
+        norm_layer: nn.Cell = nn.LayerNorm,
+        act_layer: nn.Cell = nn.GELU,
+        mlp_layer: Callable = Mlp,
+        use_rel_pos: bool = False,
+        window_size: int = 0,
+        input_size: Optional[Tuple[int, int]] = None,
+    ):
+        super().__init__()
+        self.window_size = window_size
+        self.norm1 = norm_layer((dim,))
+        self.attn = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            qk_norm=qk_norm,
+            attn_drop=attn_drop,
+            norm_layer=norm_layer,
+            use_rel_pos=use_rel_pos,
+            input_size=input_size if window_size == 0 else (window_size, window_size),
+        )
+        self.ls1 = LayerScale(dim=dim, init_values=init_values) if init_values else nn.Identity()
+        self.drop_path1 = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+
+        self.norm2 = norm_layer((dim,))
+        self.mlp = mlp_layer(
+            in_features=dim,
+            hidden_features=int(dim * mlp_ratio),
+            act_layer=act_layer,
+            drop=proj_drop,
+        )
+
+        self.ls2 = LayerScale(dim=dim, init_values=init_values) if init_values else nn.Identity()
+        self.drop_path2 = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+
+    def construct(self, x: ms.Tensor) -> ms.Tensor:
+        b, h, w, _ = x.shape
+        shortcut = x
+        x = self.norm1(x)
+
+        if self.window_size > 0:
+            # window partition
+            x, pad_hw = window_partition(x, self.window_size)
+            x = self.drop_path1(self.ls1(self.attn(x)))
+            # reverse window partition
+            x = window_unpartition(x, self.window_size, pad_hw, (h, w))
+
+        else:
+            x = self.drop_path1(self.ls1(self.attn(x)))
+
+        x = shortcut + x
+        x = ops.reshape(x, (b, h * w, -1))
+        x = x + self.drop_path2(self.ls2(self.mlp(self.norm2(x))))
+        x = ops.reshape(x, (b, h, w, -1))
+
+        return x
+
+
+class LayerScale(nn.Cell):
+    """
+    Layer scale, help ViT improve the training dynamic, allowing for the training
+    of deeper high-capacity image transformers that benefit from depth
+
+    Args:
+        dim (int): The output dimension of attnetion layer or mlp layer.
+        init_values (float): The scale factor. Default: 1e-5.
+
+    Returns:
+        Tensor, output tensor.
+
+    Examples:
+        >>> ops = LayerScale(768, 0.01)
+    """
+
+    def __init__(self, dim: int, init_values: float = 1e-5):
+        super(LayerScale, self).__init__()
+        self.gamma = Parameter(initializer(init_values, dim))
+
+    def construct(self, x):
+        return self.gamma * x
+
+
+def window_partition(x: ms.Tensor, window_size: int) -> Tuple[ms.Tensor, Tuple[int, int]]:
+    """
+    Partition into non-overlapping windows with padding if needed.
+    Args:
+        x (tensor): input tokens with [b, h, w, c].
+        window_size (int): window size.
+
+    Returns:
+        windows: windows after partition with [b * num_windows, window_size, window_size, c].
+        (hp, Wp): padded height and width before partition
+    """
+    b, h, w, c = x.shape
+
+    pad_h = (window_size - h % window_size) % window_size
+    pad_w = (window_size - w % window_size) % window_size
+
+    if pad_h > 0 or pad_w > 0:
+        # replace ops.pad with ops.concat for better performance
+        pad_mat1 = ops.zeros((b, h, pad_w, c), x.dtype)
+        pad_mat2 = ops.zeros((b, pad_h, w + pad_w, c), x.dtype)
+        x = ops.concat([ops.concat([x, pad_mat1], axis=2), pad_mat2], axis=1)
+
+    hp, wp = h + pad_h, w + pad_w
+    x = x.view(b, hp // window_size, window_size, wp // window_size, window_size, c)
+    windows = x.permute(0, 1, 3, 2, 4, 5).view(-1, window_size, window_size, c)
+
+    return windows, (hp, wp)
+
+
+def window_unpartition(
+    windows: ms.Tensor, window_size: int, pad_hw: Optional[Tuple[int, int]], hw: Tuple[int, int],
+) -> ms.Tensor:
+    """
+    Window unpartition into original sequences and removing padding.
+    Args:
+        windows (tensor): input tokens with [b * num_windows, window_size, window_size, c].
+        window_size (int): window size.
+        pad_hw (Tuple): padded height and width (hp, wp).
+        hw (Tuple): original height and width (h, w) before padding.
+
+    Returns:
+        x: unpartitioned sequences with [b, h, w, c].
+    """
+    hp, wp = pad_hw if pad_hw is not None else hw
+    h, w = hw
+    b = windows.shape[0] // (hp * wp // window_size // window_size)
+    x = windows.view(b, hp // window_size, wp // window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 3, 2, 4, 5).view(b, hp, wp, -1)
+
+    if hp > h or wp > w:
+        x = x[:, :h, :w, :]
+    return x
+
+
+class VitSAM(nn.Cell):
+    """Vision Transformer for Segment-Anything Model(SAM)"""
+
+    def __init__(
+        self,
+        image_size: int = 1024,
+        patch_size: int = 16,
+        in_chans: int = 3,
+        num_classes: int = 768,
+        embed_dim: int = 768,
+        depth: int = 12,
+        num_heads: int = 12,
+        mlp_ratio: float = 4.0,
+        qkv_bias: bool = True,
+        qk_norm: bool = False,
+        init_values: Optional[float] = None,
+        pre_norm: bool = False,
+        drop_rate: float = 0.,
+        pos_drop_rate: float = 0.,
+        proj_drop_rate: float = 0.,
+        patch_drop_rate: float = 0.,
+        attn_drop_rate: float = 0.,
+        drop_path_rate: float = 0.,
+        embed_layer: Callable = partial(PatchEmbed, output_fmt=Format.NHWC, strict_img_size=False),
+        norm_layer: Optional[Callable] = nn.LayerNorm,
+        act_layer: Optional[Callable] = nn.GELU,
+        block_fn: Callable = Block,
+        mlp_layer: Callable = Mlp,
+        use_abs_pos: bool = True,
+        use_rel_pos: bool = False,
+        window_size: int = 0,
+        global_attn_indexes: Tuple[int, ...] = (),
+        neck_chans: int = 256,
+        global_pool: str = 'avg',
+        head_hidden_size: Optional[int] = None,
+    ) -> None:
+        """
+        Args:
+            image_size (int): Input image size.
+            patch_size (int): Patch size.
+            in_chans (int): Number of input image channels.
+            embed_dim (int): Patch embedding dimension.
+            depth (int): Depth of ViT.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            init_values: Layer-scale init values (layer-scale enabled if not None).
+            drop_rate: Head dropout rate.
+            pos_drop_rate: Position embedding dropout rate.
+            attn_drop_rate: Attention dropout rate.
+            drop_path_rate: Stochastic depth rate.
+            weight_init: Weight initialization scheme.
+            embed_layer: Patch embedding layer.
+            norm_layer (nn.Cell): Normalization layer.
+            act_layer (nn.Cell): Activation layer.
+            block_fn: Transformer block layer.
+            use_abs_pos (bool): If True, use absolute positional embeddings.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            window_size (int): Window size for window attention blocks.
+            global_attn_indexes: Indexes for blocks using global attention. Used when window_size > 0.
+            global_pool: Global pooling type.
+            head_hidden_size: If set, use NormMlpHead
+        """
+        super().__init__()
+        norm_layer = norm_layer or partial(nn.LayerNorm, epsilon=1e-6)
+        act_layer = act_layer or nn.GELU
+        self.num_classes = num_classes
+        self.global_pool = global_pool
+        # num_features for consistency with other models
+        self.num_features = self.embed_dim = embed_dim
+
+        self.image_size = image_size
+        self.patch_embed = embed_layer(
+            image_size=image_size,
+            patch_size=patch_size,
+            in_chans=in_chans,
+            embed_dim=embed_dim,
+            bias=not pre_norm,
+        )
+        grid_size = self.patch_embed.grid_size
+        self.pos_embed: Optional[ms.Parameter] = None
+        if use_abs_pos:
+            # Initialize absolute positional embedding with pretrain image size.
+            self.pos_embed = ms.Parameter(ops.zeros((1, grid_size[0], grid_size[1], embed_dim)))
+        self.pos_drop = Dropout(p=pos_drop_rate)
+        if patch_drop_rate > 0:
+            self.patch_drop = PatchDropout(
+                patch_drop_rate,
+                num_prefix_tokens=self.num_prefix_tokens,
+            )
+        else:
+            self.patch_drop = nn.Identity()
+        self.norm_pre = norm_layer((embed_dim,)) if pre_norm else nn.Identity()
+        dpr = [x.item() for x in np.linspace(0, drop_path_rate, depth)]
+        self.blocks = nn.CellList([
+            block_fn(
+                dim=embed_dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_norm=qk_norm,
+                attn_drop=attn_drop_rate, proj_drop=proj_drop_rate,
+                mlp_ratio=mlp_ratio, drop_path=dpr[i], init_values=init_values,
+                act_layer=act_layer, norm_layer=norm_layer, mlp_layer=mlp_layer,
+                use_rel_pos=use_rel_pos, window_size=window_size if i not in global_attn_indexes else 0,
+                input_size=grid_size,
+            ) for i in range(depth)
+        ])
+
+        if neck_chans:
+            self.neck = nn.SequentialCell(
+                nn.Conv2d(
+                    embed_dim,
+                    neck_chans,
+                    kernel_size=1,
+                    has_bias=False,
+                ),
+                LayerNorm2d(neck_chans),
+                nn.Conv2d(neck_chans, neck_chans, kernel_size=3, padding=1, has_bias=False, pad_mode='pad'),
+                LayerNorm2d(neck_chans),
+            )
+            self.num_features = neck_chans
+        else:
+            if head_hidden_size:
+                self.neck = nn.Identity()
+            else:
+                # should have a final norm with standard ClassifierHead
+                self.neck = LayerNorm2d(embed_dim)
+            neck_chans = embed_dim
+
+        self.head = ClassifierHead(
+            neck_chans,
+            num_classes,
+            hidden_size=head_hidden_size,
+            pool_type=global_pool,
+            drop_rate=drop_rate,
+        )
+
+    def forward_features(self, x):
+        x = self.patch_embed(x)
+        if self.pos_embed is not None:
+            x = x + self.pos_embed
+        x = self.pos_drop(x)
+        x = self.patch_drop(x)
+        x = self.norm_pre(x)
+        for blk in self.blocks:
+            x = blk(x)
+        x = self.neck(x.permute(0, 3, 1, 2))
+        return x
+
+    def forward_head(self, x, pre_logits: bool = False):
+        return self.head(x, pre_logits=True) if pre_logits else self.head(x)
+
+    def construct(self, x):
+        x = self.forward_features(x)
+        x = self.forward_head(x)
+        return x
+
+
+class ClassifierHead(nn.Cell):
+    def __init__(
+        self,
+        in_features: int,
+        num_classes: int,
+        hidden_size: Optional[int] = None,
+        pool_type: str = 'avg',
+        drop_rate: float = 0.,
+        act_layer: Optional[nn.Cell] = nn.Tanh
+    ):
+        super().__init__()
+        assert pool_type == "avg", "only support avg pooling"
+        self.in_features = in_features
+        self.hidden_size = hidden_size
+        self.num_features = in_features
+        keep_dims = True if hidden_size else False
+        self.global_pool = GlobalAvgPooling(keep_dims=keep_dims)
+        self.norm = LayerNorm2d(in_features) if hidden_size else nn.Identity()
+        self.flatten = nn.Flatten() if hidden_size else nn.Identity()
+
+        if hidden_size:
+            self.pre_logits = nn.SequentialCell(OrderedDict([
+                ('fc', nn.Dense(in_features, hidden_size)),
+                ('act', act_layer()),
+            ]))
+            self.num_features = hidden_size
+        else:
+            self.pre_logits = nn.Identity()
+        self.drop = Dropout(drop_rate)
+        self.fc = nn.Dense(self.num_features, num_classes) if num_classes > 0 else nn.Identity()
+
+    def construct(self, x, pre_logits: bool = False):
+        x = self.global_pool(x)
+        x = self.norm(x)
+        x = self.flatten(x)
+        x = self.pre_logits(x)
+        x = self.drop(x)
+        if pre_logits:
+            return x
+        x = self.fc(x)
+        return x
+
+
+class LayerNorm2d(nn.Cell):
+    def __init__(self, num_channels: int, epsilon: float = 1e-6) -> None:
+        super().__init__()
+        self.weight = ms.Parameter(ops.ones(num_channels))
+        self.bias = ms.Parameter(ops.zeros(num_channels))
+        self.eps = epsilon
+
+    def construct(self, x: ms.Tensor) -> ms.Tensor:
+        u = x.mean(1, keep_dims=True)
+        s = (x - u).pow(2).mean(1, keep_dims=True)
+        x = (x - u) / ops.sqrt(s + self.eps)
+        x = self.weight[:, None, None] * x + self.bias[:, None, None]
+        return x
+
+
+@register_model
+def samvit_base_patch16(pretrained: bool = False, num_classes: int = 0, in_channels: int = 3, **kwargs):
+    default_cfg = default_cfgs["samvit_base_patch16"]
+    model = VitSAM(
+        image_size=1024, patch_size=16, embed_dim=768, depth=12, num_heads=12, global_attn_indexes=[2, 5, 8, 11],
+        window_size=14, use_rel_pos=True, num_classes=num_classes, in_chans=in_channels, **kwargs
+    )
+
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes=num_classes, in_channels=in_channels, **kwargs)
+    return model
+
+
+@register_model
+def samvit_large_patch16(pretrained: bool = False, num_classes: int = 0, in_channels: int = 3, **kwargs):
+    default_cfg = default_cfgs["samvit_large_patch16"]
+    model = VitSAM(
+        image_size=1024, patch_size=16, embed_dim=1024, depth=24, num_heads=16, global_attn_indexes=[5, 11, 17, 23],
+        window_size=14, use_rel_pos=True, num_classes=num_classes, in_chans=in_channels, **kwargs
+    )
+
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes=num_classes, in_channels=in_channels, **kwargs)
+    return model
+
+
+@register_model
+def samvit_huge_patch16(pretrained: bool = False, num_classes: int = 0, in_channels: int = 3, **kwargs):
+    default_cfg = default_cfgs["samvit_large_patch16"]
+    model = VitSAM(
+        image_size=1024, patch_size=16, embed_dim=1280, depth=32, num_heads=16, global_attn_indexes=[7, 15, 23, 31],
+        window_size=14, use_rel_pos=True, num_classes=num_classes, in_chans=in_channels, **kwargs,
+    )
+
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes=num_classes, in_channels=in_channels, **kwargs)
+    return model