diff --git a/mindone/transformers/__init__.py b/mindone/transformers/__init__.py
index a70bcee195..335da8ba1c 100644
--- a/mindone/transformers/__init__.py
+++ b/mindone/transformers/__init__.py
@@ -1087,7 +1087,10 @@
 )
 from .models.phi4_multimodal import (
     Phi4MultimodalFeatureExtractor,
+    Phi4MultimodalForCausalLM,
     Phi4MultimodalImageProcessorFast,
+    Phi4MultimodalModel,
+    Phi4MultimodalPreTrainedModel,
     Phi4MultimodalProcessor,
 )
 from .models.phimoe import PhimoeForCausalLM, PhimoeForSequenceClassification, PhimoeModel, PhimoePreTrainedModel
diff --git a/mindone/transformers/models/auto/configuration_auto.py b/mindone/transformers/models/auto/configuration_auto.py
index 80f81d9d14..e22d7b9bfb 100644
--- a/mindone/transformers/models/auto/configuration_auto.py
+++ b/mindone/transformers/models/auto/configuration_auto.py
@@ -201,6 +201,7 @@
         ("perceiver", "PerceiverConfig"),
         ("phi", "PhiConfig"),
         ("phi3", "Phi3Config"),
+        ("phi4_multimodal", "Phi4MultimodalConfig"),
         ("phimoe", "PhimoeConfig"),
         ("pix2struct", "Pix2StructConfig"),
         ("pixtral", "PixtralVisionConfig"),
@@ -473,6 +474,7 @@
         ("persimmon", "Persimmon"),
         ("phi", "Phi"),
         ("phi3", "Phi3"),
+        ("phi4_multimodal", "Phi4Multimodal"),
         ("phimoe", "Phimoe"),
         ("pegasus", "Pegasus"),
         ("pegasus_x", "PEGASUS-X"),
diff --git a/mindone/transformers/models/auto/modeling_auto.py b/mindone/transformers/models/auto/modeling_auto.py
index 57a3e36cb9..7d15c3a46d 100644
--- a/mindone/transformers/models/auto/modeling_auto.py
+++ b/mindone/transformers/models/auto/modeling_auto.py
@@ -185,6 +185,7 @@
         ("persimmon", "PersimmonModel"),
         ("phi", "PhiModel"),
         ("phi3", "Phi3Model"),
+        ("phi4_multimodal", "Phi4MultimodalModel"),
         ("pixtral", "PixtralVisionModel"),
         ("plbart", "PLBartModel"),
         ("poolformer", "PoolFormerModel"),
@@ -480,6 +481,7 @@
         ("persimmon", "PersimmonForCausalLM"),
         ("phi", "PhiForCausalLM"),
         ("phi3", "Phi3ForCausalLM"),
+        ("phi4_multimodal", "Phi4MultimodalForCausalLM"),
         ("pegasus", "PegasusForCausalLM"),
         ("plbart", "PLBartForCausalLM"),
         ("prophetnet", "ProphetNetForCausalLM"),
diff --git a/mindone/transformers/models/phi4_multimodal/__init__.py b/mindone/transformers/models/phi4_multimodal/__init__.py
index 3b4d98f65f..2e8470ef9c 100644
--- a/mindone/transformers/models/phi4_multimodal/__init__.py
+++ b/mindone/transformers/models/phi4_multimodal/__init__.py
@@ -15,6 +15,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from .modeling_phi4_multimodal import *
 from .processing_phi4_multimodal import Phi4MMAudioFeatureExtractor as Phi4MultimodalFeatureExtractor
 from .processing_phi4_multimodal import Phi4MMImageProcessor as Phi4MultimodalImageProcessorFast
 from .processing_phi4_multimodal import Phi4MMProcessor as Phi4MultimodalProcessor
diff --git a/mindone/transformers/models/phi4_multimodal/modeling_phi4_multimodal.py b/mindone/transformers/models/phi4_multimodal/modeling_phi4_multimodal.py
index 0e8a92ee0e..c82e156f64 100644
--- a/mindone/transformers/models/phi4_multimodal/modeling_phi4_multimodal.py
+++ b/mindone/transformers/models/phi4_multimodal/modeling_phi4_multimodal.py
@@ -6,6 +6,9 @@
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
 # Copyright 2025 Microsoft and the HuggingFace Inc. team. All rights reserved.
 #
+# This code is adapted from https://github.com/huggingface/transformers
+# with modifications to run transformers on mindspore.
+#
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -19,56 +22,638 @@
 # limitations under the License.
 
 import math
-from typing import Callable, Optional
+import warnings
+from typing import Callable, Optional, Union
 
 import numpy as np
 from transformers.models.phi4_multimodal.configuration_phi4_multimodal import (
     Phi4MultimodalAudioConfig,
     Phi4MultimodalConfig,
+    Phi4MultimodalVisionConfig,
 )
-from transformers.utils import auto_docstring
 
-import mindspore as ms
-import mindspore.mint.nn.functional as F
-from mindspore import Parameter, mint, nn
+import mindspore
 
 from ...activations import ACT2FN
+from ...cache_utils import Cache, DynamicCache
+from ...generation import GenerationMixin
+from ...masking_utils import create_causal_mask, create_sliding_window_causal_mask
+from ...modeling_attn_mask_utils import _prepare_4d_attention_mask
+from ...modeling_flash_attention_utils import FlashAttentionKwargs
+from ...modeling_layers import GradientCheckpointingLayer
+from ...modeling_outputs import (
+    BaseModelOutput,
+    BaseModelOutputWithPast,
+    BaseModelOutputWithPooling,
+    CausalLMOutputWithPast,
+)
+from ...modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
 from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
 from ...processing_utils import Unpack
-from ...utils.generic import TransformersKwargs
+from ...utils import can_return_tuple
+from ...utils.generic import TransformersKwargs, check_model_inputs
+
+
+class Phi4MultimodalVisionMLP(mindspore.nn.Cell):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.activation_fn = ACT2FN[config.hidden_act]
+        self.fc1 = mindspore.mint.nn.Linear(config.hidden_size, config.intermediate_size)
+        self.fc2 = mindspore.mint.nn.Linear(config.intermediate_size, config.hidden_size)
+
+    def construct(self, hidden_states: mindspore.Tensor) -> mindspore.Tensor:
+        hidden_states = self.fc1(hidden_states)
+        hidden_states = self.activation_fn(hidden_states)
+        hidden_states = self.fc2(hidden_states)
+        return hidden_states
 
 
 def simple_eager_attention_forward(
-    module: nn.Cell,
-    query_states: ms.Tensor,
-    key_states: ms.Tensor,
-    value_states: ms.Tensor,
-    attention_mask: Optional[ms.Tensor],
+    module: mindspore.nn.Cell,
+    query_states: mindspore.Tensor,
+    key_states: mindspore.Tensor,
+    value_states: mindspore.Tensor,
+    attention_mask: Optional[mindspore.Tensor],
     scaling: float,
     dropout: float = 0.0,
     **kwargs: Unpack[TransformersKwargs],
 ):
-    attn_weights = mint.matmul(query_states, key_states.transpose(2, 3)) * scaling
+    attn_weights = mindspore.mint.matmul(query_states, key_states.transpose(2, 3)) * scaling
     if attention_mask is not None:
         causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
         attn_weights = attn_weights + causal_mask
 
-    attn_weights = mint.nn.functional.softmax(attn_weights, dim=-1, dtype=ms.float32).to(query_states.dtype)
-    attn_weights = mint.nn.functional.dropout(attn_weights, p=dropout, training=module.training)
-    attn_output = mint.matmul(attn_weights, value_states)
+    attn_weights = mindspore.mint.nn.functional.softmax(attn_weights, dim=-1, dtype=mindspore.float32).to(
+        query_states.dtype
+    )
+    attn_weights = mindspore.mint.nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = mindspore.mint.matmul(attn_weights, value_states)
     attn_output = attn_output.transpose(1, 2).contiguous()
 
     return attn_output, attn_weights
 
 
-class Phi4MultimodalAudioMLP(nn.Cell):
+class Phi4MultimodalVisionAttention(mindspore.nn.Cell):
+    def __init__(self, config: Phi4MultimodalVisionConfig):
+        super().__init__()
+        self.config = config
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_heads
+        self.scaling = self.head_dim**-0.5
+        self.is_causal = True
+        self.attention_dropout = config.attention_dropout
+
+        self.k_proj = mindspore.mint.nn.Linear(config.hidden_size, config.hidden_size)
+        self.v_proj = mindspore.mint.nn.Linear(config.hidden_size, config.hidden_size)
+        self.q_proj = mindspore.mint.nn.Linear(config.hidden_size, config.hidden_size)
+        self.out_proj = mindspore.mint.nn.Linear(config.hidden_size, config.hidden_size)
+
+    def construct(
+        self,
+        hidden_states: mindspore.Tensor,
+        attention_mask: Optional[mindspore.Tensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple[mindspore.Tensor, Optional[mindspore.Tensor]]:
+        """Input shape: Batch x Time x Channel"""
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+
+        attention_interface: Callable = simple_eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1)
+        attn_output = self.out_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class Phi4MultimodalVisionEncoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: Phi4MultimodalVisionConfig):
+        super().__init__()
+        self.embed_dim = config.hidden_size
+        self.layer_norm1 = mindspore.mint.nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
+        self.self_attn = Phi4MultimodalVisionAttention(config)
+        self.layer_norm2 = mindspore.mint.nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
+        self.mlp = Phi4MultimodalVisionMLP(config)
+
+    def construct(
+        self,
+        hidden_states: mindspore.Tensor,
+        attention_mask: mindspore.Tensor,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> mindspore.Tensor:
+        residual = hidden_states
+
+        hidden_states = self.layer_norm1(hidden_states)
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        residual = hidden_states
+        hidden_states = self.layer_norm2(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        return hidden_states
+
+
+class Phi4MultimodalVisionEncoder(mindspore.nn.Cell):
+    """
+    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
+    [`Phi4MultimodalVisionEncoderLayer`].
+
+    Args:
+        config: Phi4MultimodalVisionConfig
+    """
+
+    def __init__(self, config: Phi4MultimodalVisionConfig):
+        super().__init__()
+        self.config = config
+        self.layers = mindspore.nn.CellList(
+            [Phi4MultimodalVisionEncoderLayer(config) for _ in range(config.num_hidden_layers)]
+        )
+        self.gradient_checkpointing = False
+
+    # Ignore copy
+
+    def construct(
+        self,
+        inputs_embeds,
+        attention_mask: Optional[mindspore.Tensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutput:
+        hidden_states = inputs_embeds
+        for encoder_layer in self.layers:
+            hidden_states = encoder_layer(
+                hidden_states,
+                attention_mask,
+                **kwargs,
+            )
+
+        return BaseModelOutput(last_hidden_state=hidden_states)
+
+
+def _trunc_normal_(tensor, mean, std, a, b):
+    # Cut & paste from PyTorch official master until it's in a few official releases - RW
+    # Method based on https://people.sc.fsu.edu/~jburkardt/presentations/truncated_normal.pdf
+    def norm_cdf(x):
+        # Computes standard normal cumulative distribution function
+        return (1.0 + math.erf(x / math.sqrt(2.0))) / 2.0
+
+    if (mean < a - 2 * std) or (mean > b + 2 * std):
+        warnings.warn(
+            "mean is more than 2 std from [a, b] in nn.init.trunc_normal_. "
+            "The distribution of values may be incorrect.",
+            stacklevel=2,
+        )
+
+    # Values are generated by using a truncated uniform distribution and
+    # then using the inverse CDF for the normal distribution.
+    # Get upper and lower cdf values
+    value_l = norm_cdf((a - mean) / std)
+    u = norm_cdf((b - mean) / std)
+
+    # Uniformly fill tensor with values from [l, u], then translate to
+    # [2l-1, 2u-1].
+    tensor.uniform_(2 * value_l - 1, 2 * u - 1)
+
+    # Use inverse cdf transform for normal distribution to get truncated
+    # standard normal
+    tensor.erfinv_()
+
+    # Transform to proper mean, std
+    tensor.mul_(std * math.sqrt(2.0))
+    tensor.add_(mean)
+
+    # Clamp to ensure it's in the proper range
+    tensor.clamp_(min=a, max=b)
+
+
+def trunc_normal_tf_(
+    tensor: mindspore.Tensor, mean: float = 0.0, std: float = 1.0, a: float = -2.0, b: float = 2.0
+) -> mindspore.Tensor:
+    """Fills the input Tensor with values drawn from a truncated
+    normal distribution. The values are effectively drawn from the
+    normal distribution :math:`\\mathcal{N}(\text{mean}, \text{std}^2)`
+    with values outside :math:`[a, b]` redrawn until they are within
+    the bounds. The method used for generating the random values works
+    best when :math:`a \\leq \text{mean} \\leq b`.
+
+    NOTE: this 'tf' variant behaves closer to Tensorflow / JAX impl where the
+    bounds [a, b] are applied when sampling the normal distribution with mean=0, std=1.0
+    and the result is subsequently scaled and shifted by the mean and std args.
+
+    Args:
+        tensor: an n-dimensional `mindspore.Tensor`
+        mean: the mean of the normal distribution
+        std: the standard deviation of the normal distribution
+        a: the minimum cutoff value
+        b: the maximum cutoff value
+    """
+    _trunc_normal_(tensor, 0, 1.0, a, b)
+    tensor.mul_(std).add_(mean)
+
+
+def variance_scaling_(tensor, scale=1.0, mode="fan_in", distribution="normal"):
+    fan_in, fan_out = 0.5, 0.5  # _calculate_fan_in_and_fan_out(tensor)
+    if mode == "fan_in":
+        denom = fan_in
+    elif mode == "fan_out":
+        denom = fan_out
+    elif mode == "fan_avg":
+        denom = (fan_in + fan_out) / 2
+
+    variance = scale / denom
+
+    if distribution == "truncated_normal":
+        # constant is stddev of standard normal truncated to (-2, 2)
+        trunc_normal_tf_(tensor, std=math.sqrt(variance) / 0.87962566103423978)
+    elif distribution == "normal":
+        with mindspore._no_grad():
+            tensor.normal_(std=math.sqrt(variance))
+    elif distribution == "uniform":
+        bound = math.sqrt(3 * variance)
+        with mindspore._no_grad():
+            tensor.uniform_(-bound, bound)
+    else:
+        raise ValueError(f"invalid distribution {distribution}")
+
+
+def lecun_normal_(tensor):
+    variance_scaling_(tensor, mode="fan_in", distribution="truncated_normal")
+
+
+def default_flax_embed_init(tensor):
+    variance_scaling_(tensor, mode="fan_in", distribution="normal")
+
+
+class Phi4MultimodalVisionPreTrainedModel(PreTrainedModel):
+    config: Phi4MultimodalVisionConfig
+    base_model_prefix = "phi4_vision"
+    supports_gradient_checkpointing = True
+
+    _no_split_modules = ["Phi4MultimodalVisionEncoderLayer"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _supports_attention_backend = True
+
+    _can_record_outputs = {
+        "hidden_states": Phi4MultimodalVisionEncoderLayer,
+        "attentions": Phi4MultimodalVisionAttention,
+    }
+
+    def _init_weights(self, module):
+        """Initialize the weights"""
+        pass
+
+
+class Phi4MultimodalVisionEmbeddings(mindspore.nn.Cell):
+    def __init__(self, config: Phi4MultimodalVisionConfig):
+        super().__init__()
+        self.config = config
+        self.patch_size = config.patch_size
+        self.num_patches_per_side = config.image_size // self.patch_size
+
+        self.patch_embedding = mindspore.mint.nn.Conv2d(
+            in_channels=config.num_channels,
+            out_channels=config.hidden_size,
+            kernel_size=self.patch_size,
+            stride=self.patch_size,
+            padding="valid",
+        )
+        self.position_embedding = mindspore.mint.nn.Embedding(self.num_patches_per_side**2, config.hidden_size)
+
+    def interpolate_pos_encoding(self, embeddings: mindspore.Tensor, height: int, width: int) -> mindspore.Tensor:
+        """
+        This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher resolution
+        images.
+
+        Adapted from:
+        - https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174-L194, and
+        - https://github.com/facebookresearch/dinov2/blob/e1277af2ba9496fbadf7aec6eba56e8d882d1e35/dinov2/models/vision_transformer.py#L179-L211
+        """
+
+        num_patches = embeddings.shape[1]
+        num_positions = self.position_embedding.weight.shape[0]
+
+        # always interpolate when tracing to ensure the exported model works for dynamic input shapes
+        if num_patches == num_positions and height == width:
+            return self.position_embedding(self.position_ids)
+
+        patch_pos_embed = self.position_embedding.weight.unsqueeze(0)
+
+        dim = embeddings.shape[-1]
+
+        new_height = height // self.patch_size
+        new_width = width // self.patch_size
+
+        sqrt_num_positions = int(num_positions**0.5)
+        patch_pos_embed = patch_pos_embed.reshape(1, sqrt_num_positions, sqrt_num_positions, dim)
+        patch_pos_embed = patch_pos_embed.permute(0, 3, 1, 2)
+
+        patch_pos_embed = mindspore.mint.nn.functional.interpolate(
+            patch_pos_embed,
+            size=(new_height, new_width),
+            mode="bicubic",
+            align_corners=False,
+        )
+
+        patch_pos_embed = patch_pos_embed.permute(0, 2, 3, 1).view(1, -1, dim)
+        return patch_pos_embed
+
+    def construct(self, pixel_values: mindspore.Tensor, patch_attention_mask: mindspore.Tensor) -> mindspore.Tensor:
+        batch_size = pixel_values.shape[0]
+
+        patch_embeds = self.patch_embedding(pixel_values)
+        embeddings = patch_embeds.flatten(2).transpose(1, 2)
+
+        max_im_h, max_im_w = pixel_values.shape[2], pixel_values.shape[3]
+        max_nb_patches_h, max_nb_patches_w = max_im_h // self.patch_size, max_im_w // self.patch_size
+        boundaries = mindspore.mint.arange(1 / self.num_patches_per_side, 1.0, 1 / self.num_patches_per_side)
+        position_ids = mindspore.ops.full((batch_size, max_nb_patches_h * max_nb_patches_w), fill_value=0)
+
+        for batch_idx, p_attn_mask in enumerate(patch_attention_mask):
+            nb_patches_h = p_attn_mask[:, 0].sum()
+            nb_patches_w = p_attn_mask[0].sum()
+
+            fractional_coords_h = mindspore.mint.arange(0, 1 - 1e-6, 1 / nb_patches_h)
+            fractional_coords_w = mindspore.mint.arange(0, 1 - 1e-6, 1 / nb_patches_w)
+
+            bucket_coords_h = mindspore.ops.bucketize(fractional_coords_h, boundaries, right=True)
+            bucket_coords_w = mindspore.ops.bucketize(fractional_coords_w, boundaries, right=True)
+
+            pos_ids = (bucket_coords_h[:, None] * self.num_patches_per_side + bucket_coords_w).flatten()
+            position_ids[batch_idx][p_attn_mask.view(-1).cpu()] = pos_ids
+
+        embeddings = embeddings + self.position_embedding(position_ids)
+        return embeddings
+
+
+class Phi4MultimodalVisionMultiheadAttentionPoolingHead(mindspore.nn.Cell):
+    """Multihead Attention Pooling."""
+
+    def __init__(self, config: Phi4MultimodalVisionConfig):
+        super().__init__()
+
+        self.probe = mindspore.Parameter(mindspore.mint.randn(1, 1, config.hidden_size))
+        self.attention = mindspore.nn.MultiheadAttention(
+            config.hidden_size, config.num_attention_heads, batch_first=True
+        )
+        self.layernorm = mindspore.mint.nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.mlp = Phi4MultimodalVisionMLP(config)
+
+    def construct(self, hidden_state, attention_mask):
+        batch_size = hidden_state.shape[0]
+        probe = self.probe.repeat(batch_size, 1, 1)
+
+        hidden_state = self.attention(
+            query=probe, key=hidden_state, value=hidden_state, key_padding_mask=~attention_mask
+        )[0]
+
+        residual = hidden_state
+        hidden_state = self.layernorm(hidden_state)
+        hidden_state = residual + self.mlp(hidden_state)
+
+        return hidden_state[:, 0]
+
+
+class Phi4MultimodalVisionModel(Phi4MultimodalVisionPreTrainedModel):
+    config: Phi4MultimodalVisionConfig
+    main_input_name = "pixel_values"
+
+    def __init__(self, config: Phi4MultimodalVisionConfig):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = Phi4MultimodalVisionEmbeddings(config)
+        self.encoder = Phi4MultimodalVisionEncoder(config)
+        self.post_layernorm = mindspore.mint.nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.head = Phi4MultimodalVisionMultiheadAttentionPoolingHead(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self) -> mindspore.nn.Cell:
+        return self.embeddings.patch_embedding
+
+    @check_model_inputs
+    def construct(
+        self,
+        pixel_values,
+        patch_attention_mask: Optional[mindspore.Tensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutputWithPooling:
+        batch_size = pixel_values.shape[0]
+        if patch_attention_mask is None:
+            patch_attention_mask = mindspore.mint.ones(
+                size=(
+                    batch_size,
+                    pixel_values.shape[2] // self.config.patch_size,
+                    pixel_values.shape[3] // self.config.patch_size,
+                ),
+                dtype=mindspore.bool_,
+            )
+
+        hidden_states = self.embeddings(pixel_values=pixel_values, patch_attention_mask=patch_attention_mask)
+
+        patch_attention_mask = patch_attention_mask.view(batch_size, -1)
+        # The call to `_upad_input` in `_flash_attention_forward` is expensive
+        # So when the `patch_attention_mask` is full of 1s (i.e. attending to the whole sequence),
+        # avoiding passing the attention_mask, which is equivalent to attending to the full sequence
+        if not mindspore.mint.any(~patch_attention_mask):
+            attention_mask = None
+        else:
+            attention_mask = (
+                _prepare_4d_attention_mask(patch_attention_mask, hidden_states.dtype)
+                if self.config._attn_implementation != "flash_attention_2"
+                else patch_attention_mask
+            )
+
+        encoder_outputs: BaseModelOutput = self.encoder(
+            inputs_embeds=hidden_states,
+            attention_mask=attention_mask,
+            **kwargs,
+        )
+
+        last_hidden_state = encoder_outputs.last_hidden_state
+        last_hidden_state = self.post_layernorm(last_hidden_state)
+
+        pooled_output = self.head(
+            hidden_state=last_hidden_state,
+            attention_mask=patch_attention_mask,
+        )
+
+        return BaseModelOutputWithPooling(
+            last_hidden_state=last_hidden_state,
+            pooler_output=pooled_output,
+        )
+
+
+class Phi4MultimodalImageEmbedding(mindspore.nn.Cell):
+    """Image embedding."""
+
+    def __init__(self, config: Phi4MultimodalConfig):
+        super().__init__()
+        self.config = config
+        self.layer_idx = config.vision_config.feature_layer
+        self.crop_size = config.vision_config.crop_size
+        self.image_dim_out = config.vision_config.hidden_size
+
+        n_patches = config.vision_config.image_size // config.vision_config.patch_size
+        if n_patches % 2 != 0:
+            self.img_processor_padding = mindspore.mint.nn.ReflectionPad2d((0, 1, 0, 1))
+            n_patches += 1
+        self.num_img_tokens = (n_patches // 2) ** 2
+
+        self.drop = mindspore.mint.nn.Dropout(config.embd_pdrop)
+        self.img_processor = Phi4MultimodalVisionModel._from_config(config.vision_config)
+        self.image_token_compression = mindspore.mint.nn.AvgPool2d(kernel_size=2, stride=2)
+        self.img_projection_up = mindspore.mint.nn.Linear(self.image_dim_out, config.hidden_size)
+        self.img_projection_down = mindspore.mint.nn.Linear(config.hidden_size, config.hidden_size)
+        self.global_img_feature_extensor = mindspore.Parameter(mindspore.mint.zeros([1, 1, self.image_dim_out]))
+        self.sub_img_feature_extensor = mindspore.Parameter(mindspore.mint.zeros([1, 1, 1, self.image_dim_out]))
+
+    def get_img_features(self, img_embeds: mindspore.Tensor, attention_mask=None) -> mindspore.Tensor:
+        img_processor_output = self.img_processor(
+            img_embeds, patch_attention_mask=attention_mask, output_hidden_states=True
+        )
+        img_feature = img_processor_output.hidden_states[self.layer_idx]
+
+        patch_feature = img_feature
+        # reshape to 2D tensor
+        width = int(math.sqrt(patch_feature.shape[1]))
+        patch_feature = patch_feature.view(-1, width, width, patch_feature.shape[-1])
+        # convert to NCHW
+        patch_feature = patch_feature.permute(0, 3, 1, 2)
+        if getattr(self, "img_processor_padding", None) is not None:
+            patch_feature = self.img_processor_padding(patch_feature)
+        patch_feature = self.image_token_compression(patch_feature)
+        # convert to NHWC
+        patch_feature = patch_feature.permute(0, 2, 3, 1)
+        patch_feature = patch_feature.view(-1, patch_feature.shape[1] * patch_feature.shape[2], patch_feature.shape[-1])
+        return patch_feature
+
+    def construct(
+        self,
+        input_ids: mindspore.Tensor,
+        inputs_embeds: mindspore.Tensor,
+        image_pixel_values: mindspore.Tensor,
+        image_sizes: Optional[mindspore.Tensor] = None,
+        image_attention_mask: Optional[mindspore.Tensor] = None,
+    ) -> mindspore.Tensor:
+        image_pixel_values = image_pixel_values.to(self.img_processor.embeddings.patch_embedding.weight.dtype)
+
+        target_dtype = self.img_projection_up.bias.dtype
+
+        batch_size = image_pixel_values.shape[0]
+
+        img_features = self.get_img_features(
+            image_pixel_values.flatten(0, 1),
+            attention_mask=image_attention_mask.flatten(0, 1).to(
+                dtype=bool,
+            ),
+        )
+        base_feat_size = int(np.sqrt(img_features.shape[1]))
+        img_features = img_features.view(batch_size, -1, base_feat_size**2, self.image_dim_out)
+        image_sizes = image_sizes.view(-1, 2)
+
+        output_imgs = []
+        for idx in range(batch_size):
+            height, width = image_sizes[idx]
+            height_ratio = height // self.crop_size
+            width_ratio = width // self.crop_size
+            area_ratio = height_ratio * width_ratio
+
+            global_img = img_features[idx, :1]
+            global_img = global_img.reshape(1, base_feat_size, base_feat_size, self.image_dim_out).contiguous()
+            temporary_extensor = self.sub_img_feature_extensor.repeat(1, base_feat_size, 1, 1)
+            global_img = mindspore.mint.cat([global_img, temporary_extensor], dim=2).reshape(1, -1, self.image_dim_out)
+
+            sub_img = img_features[idx, 1:]
+            sub_img = sub_img[:area_ratio]
+            sub_img = (
+                sub_img.reshape(height_ratio, width_ratio, base_feat_size, base_feat_size, self.image_dim_out)
+                .transpose(1, 2)
+                .reshape(1, height_ratio * base_feat_size, width_ratio * base_feat_size, self.image_dim_out)
+                .contiguous()
+            )
+
+            if image_attention_mask is not None:
+                reshaped_image_attention_mask = (
+                    image_attention_mask[idx, 1 : area_ratio + 1, 0::2, 0::2]
+                    .reshape(height_ratio, width_ratio, base_feat_size, base_feat_size)
+                    .transpose(1, 2)
+                    .reshape(1, height_ratio * base_feat_size, width_ratio * base_feat_size)
+                )
+                useful_height = int(reshaped_image_attention_mask[0, :, 0].sum().item())
+                useful_width = int(reshaped_image_attention_mask[0, 0, :].sum().item())
+                sub_img = sub_img[:, :useful_height, :useful_width]
+                temporary_extensor = self.sub_img_feature_extensor.repeat(1, useful_height, 1, 1)
+            else:
+                temporary_extensor = self.sub_img_feature_extensor.repeat(1, height_ratio * base_feat_size, 1, 1)
+
+            sub_img = mindspore.mint.cat([sub_img, temporary_extensor], dim=2).reshape(1, -1, self.image_dim_out)
+
+            # Merge global and sub
+            output_imgs.append(mindspore.mint.cat([sub_img, self.global_img_feature_extensor, global_img], dim=1))
+
+        img_set_tensor = []
+        for output_img in output_imgs:
+            output_img = output_img.to(dtype=target_dtype)
+            img_feature_proj = self.img_projection_up(output_img)
+            img_feature_proj = mindspore.mint.nn.functional.gelu(img_feature_proj)
+            img_feature_proj = self.img_projection_down(img_feature_proj)
+            img_set_tensor.append(img_feature_proj)
+
+        merged_img_set_tensor = mindspore.mint.cat(img_set_tensor, dim=1).squeeze(0)
+        merged_img_set_tensor = merged_img_set_tensor.to(
+            dtype=inputs_embeds.dtype,
+        )
+
+        with mindspore._no_grad():
+            positions_tuple = mindspore.mint.nonzero(
+                input_ids == self.config.vision_config.image_token_id, as_tuple=True
+            )
+
+        # Temporarily disable autocast to avoid issue on bf16 tensors
+        # Ref: https://github.com/pytorch/pytorch/issues/132715
+        image_embeds = inputs_embeds.index_put(indices=positions_tuple, values=merged_img_set_tensor, accumulate=False)
+
+        image_embeds = self.drop(image_embeds)
+
+        return image_embeds
+
+
+# AUDIO
+class Phi4MultimodalAudioMLP(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig):
         super().__init__()
-        self.layer_norm = mint.nn.LayerNorm(config.hidden_size)
+        self.layer_norm = mindspore.mint.nn.LayerNorm(config.hidden_size)
         self.act_fn = ACT2FN[config.activation]
-        self.gate_up_proj = mint.nn.Linear(config.hidden_size, config.intermediate_size * 2)
-        self.down_proj = mint.nn.Linear(config.intermediate_size, config.hidden_size)
-        self.dropout = nn.Dropout(config.dropout_rate)
+        self.gate_up_proj = mindspore.mint.nn.Linear(config.hidden_size, config.intermediate_size * 2)
+        self.down_proj = mindspore.mint.nn.Linear(config.intermediate_size, config.hidden_size)
+        self.dropout = mindspore.mint.nn.Dropout(config.dropout_rate)
 
     def construct(self, hidden_states):
         hidden_states = self.layer_norm(hidden_states)
@@ -82,7 +667,7 @@ def construct(self, hidden_states):
         return out
 
 
-class Phi4MultimodalAudioAttention(nn.Cell):
+class Phi4MultimodalAudioAttention(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig):
         super().__init__()
         self.config = config
@@ -91,15 +676,23 @@ def __init__(self, config: Phi4MultimodalAudioConfig):
         self.attention_dropout = config.dropout_rate
         self.is_causal = True
 
-        self.q_proj = mint.nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=True)
-        self.k_proj = mint.nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=True)
-        self.v_proj = mint.nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=True)
-        self.o_proj = mint.nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=True)
+        self.q_proj = mindspore.mint.nn.Linear(
+            config.hidden_size, config.num_attention_heads * self.head_dim, bias=True
+        )
+        self.k_proj = mindspore.mint.nn.Linear(
+            config.hidden_size, config.num_attention_heads * self.head_dim, bias=True
+        )
+        self.v_proj = mindspore.mint.nn.Linear(
+            config.hidden_size, config.num_attention_heads * self.head_dim, bias=True
+        )
+        self.o_proj = mindspore.mint.nn.Linear(
+            config.num_attention_heads * self.head_dim, config.hidden_size, bias=True
+        )
 
     def construct(
         self,
-        hidden_states: ms.Tensor,
-        attention_mask: ms.Tensor,
+        hidden_states: mindspore.Tensor,
+        attention_mask: mindspore.Tensor,
         **kwargs,
     ):
         input_shape = hidden_states.shape[:-1]
@@ -129,10 +722,10 @@ def construct(
         return attn_output
 
 
-class Phi4MultimodalAudioDepthWiseSeparableConv1d(nn.Cell):
+class Phi4MultimodalAudioDepthWiseSeparableConv1d(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig, padding: int = 0):
         super().__init__()
-        self.dw_conv = mint.nn.Conv1d(
+        self.dw_conv = mindspore.mint.nn.Conv1d(
             config.hidden_size,
             config.hidden_size * config.depthwise_multiplier,
             config.kernel_size,
@@ -140,7 +733,7 @@ def __init__(self, config: Phi4MultimodalAudioConfig, padding: int = 0):
             padding=padding,
             groups=config.hidden_size,
         )
-        self.pw_conv = mint.nn.Conv1d(
+        self.pw_conv = mindspore.mint.nn.Conv1d(
             config.hidden_size * config.depthwise_multiplier, config.depthwise_separable_out_channel, 1, 1, 0
         )
 
@@ -148,16 +741,18 @@ def construct(self, hidden_states):
         return self.pw_conv(self.dw_conv(hidden_states))
 
 
-class Phi4MultimodalAudioGluPointWiseConv(nn.Cell):
+class Phi4MultimodalAudioGluPointWiseConv(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig):
         super().__init__()
         self.config = config
         self.output_dim = config.ext_pw_out_channel
 
-        self.ext_pw_conv_1d = mint.nn.Conv1d(config.hidden_size, config.ext_pw_out_channel * 2, kernel_size=1, stride=1)
+        self.ext_pw_conv_1d = mindspore.mint.nn.Conv1d(
+            config.hidden_size, config.ext_pw_out_channel * 2, kernel_size=1, stride=1
+        )
         self.glu_act = ACT2FN[config.conv_glu_type]
-        self.b1 = Parameter(mint.zeros((1, config.ext_pw_out_channel, 1)))
-        self.b2 = Parameter(mint.zeros((1, config.ext_pw_out_channel, 1)))
+        self.b1 = mindspore.Parameter(mindspore.mint.zeros((1, config.ext_pw_out_channel, 1)))
+        self.b2 = mindspore.Parameter(mindspore.mint.zeros((1, config.ext_pw_out_channel, 1)))
 
     def construct(self, hidden_states):
         # we assume the input always has the #channel (#dim) in the last dimension of the
@@ -169,20 +764,22 @@ def construct(self, hidden_states):
         return out.permute([0, 2, 1])
 
 
-class Phi4MultimodalAudioConvModule(nn.Cell):
+class Phi4MultimodalAudioConvModule(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig):
         super().__init__()
         self.config = config
         self.kernel_size = config.kernel_size
 
-        self.layer_norm = mint.nn.LayerNorm(config.hidden_size)
+        self.layer_norm = mindspore.mint.nn.LayerNorm(config.hidden_size)
         self.glu = Phi4MultimodalAudioGluPointWiseConv(config)
         self.dw_sep_conv_1d = Phi4MultimodalAudioDepthWiseSeparableConv1d(config, padding=config.kernel_size - 1)
         self.act = ACT2FN[config.conv_activation]
-        self.ext_pw_conv_1d = mint.nn.Conv1d(config.hidden_size, config.ext_pw_out_channel, kernel_size=1, stride=1)
-        self.dropout = nn.Dropout(config.dropout_rate)
+        self.ext_pw_conv_1d = mindspore.mint.nn.Conv1d(
+            config.hidden_size, config.ext_pw_out_channel, kernel_size=1, stride=1
+        )
+        self.dropout = mindspore.mint.nn.Dropout(config.dropout_rate)
 
-    def construct(self, hidden_states: ms.Tensor):
+    def construct(self, hidden_states: mindspore.Tensor):
         hidden_states = self.glu(self.layer_norm(hidden_states))
         hidden_states = self.dw_sep_conv_1d(hidden_states.permute([0, 2, 1]))
 
@@ -195,7 +792,7 @@ def construct(self, hidden_states: ms.Tensor):
         return out
 
 
-class Phi4MultimodalAudioConformerEncoderLayer(nn.Cell):
+class Phi4MultimodalAudioConformerEncoderLayer(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig):
         super().__init__()
 
@@ -203,13 +800,13 @@ def __init__(self, config: Phi4MultimodalAudioConfig):
         self.self_attn = Phi4MultimodalAudioAttention(config)
         self.conv = Phi4MultimodalAudioConvModule(config)
         self.feed_forward_out = Phi4MultimodalAudioMLP(config)
-        self.layer_norm_att = mint.nn.LayerNorm(config.hidden_size)
-        self.layer_norm = mint.nn.LayerNorm(config.hidden_size)
+        self.layer_norm_att = mindspore.mint.nn.LayerNorm(config.hidden_size)
+        self.layer_norm = mindspore.mint.nn.LayerNorm(config.hidden_size)
 
     def construct(
         self,
-        hidden_states: ms.Tensor,
-        attention_mask: ms.Tensor,
+        hidden_states: mindspore.Tensor,
+        attention_mask: mindspore.Tensor,
     ):
         residual = hidden_states + 0.5 * self.feed_forward_in(hidden_states)
         hidden_states = self.layer_norm_att(residual)
@@ -223,7 +820,7 @@ def construct(
         return out
 
 
-class Phi4MultimodalAudioNemoConvSubsampling(nn.Cell):
+class Phi4MultimodalAudioNemoConvSubsampling(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig):
         super().__init__()
         self.subsampling_factor = config.time_reduction
@@ -232,31 +829,33 @@ def __init__(self, config: Phi4MultimodalAudioConfig):
         conv_channels = config.nemo_conv_channels
 
         layers = [
-            mint.nn.Conv2d(1, conv_channels, kernel_size=3, stride=2, padding=1),
+            mindspore.mint.nn.Conv2d(1, conv_channels, kernel_size=3, stride=2, padding=1),
             self.act_fn,
         ]
         for _ in range(self.sampling_num - 1):
             layers.extend(
                 [
-                    mint.nn.Conv2d(
+                    mindspore.mint.nn.Conv2d(
                         conv_channels, conv_channels, kernel_size=3, stride=2, padding=1, groups=conv_channels
                     ),
-                    mint.nn.Conv2d(conv_channels, conv_channels, kernel_size=1, stride=1, padding=0, groups=1),
+                    mindspore.mint.nn.Conv2d(
+                        conv_channels, conv_channels, kernel_size=1, stride=1, padding=0, groups=1
+                    ),
                     self.act_fn,
                 ]
             )
 
         # Aggregate the layers
-        self.conv = ms.nn.SequentialCell(*layers)
-        self.out = mint.nn.Linear(conv_channels * config.nemo_final_size, config.hidden_size)
+        self.conv = mindspore.nn.SequentialCell(*layers)
+        self.out = mindspore.mint.nn.Linear(conv_channels * config.nemo_final_size, config.hidden_size)
 
-    def construct(self, hidden_states: ms.Tensor, mask: Optional[ms.Tensor]):
+    def construct(self, hidden_states: mindspore.Tensor, mask: Optional[mindspore.Tensor]):
         # Unsqueeze Channel Axis
         hidden_states = hidden_states.unsqueeze(1)
         hidden_states = self.conv(hidden_states)
 
         # Flatten Channel and Frequency Axes
-        b, _, t, _ = hidden_states.size()
+        b, _, t, _ = hidden_states.shape
         hidden_states = self.out(hidden_states.transpose(1, 2).reshape(b, t, -1))
 
         if mask is None:
@@ -264,13 +863,16 @@ def construct(self, hidden_states: ms.Tensor, mask: Optional[ms.Tensor]):
 
         max_audio_length = hidden_states.shape[1]
         feature_lens = mask.sum(1)
-        padding_length = mint.ceil(feature_lens / self.subsampling_factor)
-        arange_ = mint.arange(0, max_audio_length)
-        pad_mask = arange_.expand(padding_length.size(0), -1) < padding_length.unsqueeze(1)
+        padding_length = mindspore.mint.ceil(feature_lens / self.subsampling_factor)
+        arange_ = mindspore.mint.arange(
+            0,
+            max_audio_length,
+        )
+        pad_mask = arange_.expand(padding_length.shape[0], -1) < padding_length.unsqueeze(1)
         return hidden_states, pad_mask.unsqueeze(1)
 
 
-class Phi4MultimodalAudioRelativeAttentionBias(nn.Cell):
+class Phi4MultimodalAudioRelativeAttentionBias(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig):
         super().__init__()
 
@@ -279,13 +881,13 @@ def __init__(self, config: Phi4MultimodalAudioConfig):
         self.num_buckets = self.max_distance
         if not config.bias_symmetric:
             self.num_buckets *= 2
-        self.bias_values = mint.nn.Embedding(self.num_buckets, config.num_attention_heads)
+        self.bias_values = mindspore.mint.nn.Embedding(self.num_buckets, config.num_attention_heads)
 
     def construct(self, x):
         # instantiate bias compatible with shape of x
         max_pos = x.shape[1]
-        context_position = mint.arange(max_pos, dtype=ms.int64)[:, None]
-        memory_position = mint.arange(max_pos, dtype=ms.int64)[None, :]
+        context_position = mindspore.mint.arange(max_pos, dtype=mindspore.int64)[:, None]
+        memory_position = mindspore.mint.arange(max_pos, dtype=mindspore.int64)[None, :]
         relative_position = memory_position - context_position
         # clipping to a maximum distance using ops that play well with ONNX export
         relative_position = relative_position.masked_fill(relative_position < -self.max_distance, -self.max_distance)
@@ -303,17 +905,16 @@ def construct(self, x):
         return att_bias
 
 
-class Phi4MultimodalAudioMeanVarianceNormLayer(nn.Cell):
+class Phi4MultimodalAudioMeanVarianceNormLayer(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalAudioConfig):
         super().__init__()
-        self.register_buffer("global_mean", mint.zeros(config.input_size))
-        self.register_buffer("global_invstd", mint.ones(config.input_size))
+        self.register_buffer("global_mean", mindspore.mint.zeros(config.input_size))
+        self.register_buffer("global_invstd", mindspore.mint.ones(config.input_size))
 
     def construct(self, x):
         return (x - self.global_mean) * self.global_invstd
 
 
-@auto_docstring
 class Phi4MultimodalAudioPreTrainedModel(PreTrainedModel):
     config: Phi4MultimodalAudioConfig
     supports_gradient_checkpointing = True
@@ -339,7 +940,9 @@ def unfold_tensor(tensor, max_seq_len):
     _, _, D = tensor.shape
     tensor = tensor.transpose(-1, -2)
     # N x D x 1 x T => N x (D x max_seq_len) x T'
-    tensor = F.unfold(tensor[..., None, :], kernel_size=(1, max_seq_len), stride=(1, max_seq_len))
+    tensor = mindspore.mint.nn.functional.unfold(
+        tensor[..., None, :], kernel_size=(1, max_seq_len), stride=(1, max_seq_len)
+    )
 
     new_bsz, _, slen = tensor.shape
     tensor = tensor.view(new_bsz, -1, max_seq_len, slen)
@@ -357,18 +960,18 @@ def adaptive_enc_mask(x_len, chunk_start_idx, left_window=0, right_window=0):
         left_window (int): how many left chunks can be seen
         right_window (int): how many right chunks can be seen. It is used for chunk overlap model.
         Returns:
-            mask (ms.Tensor): a mask tensor for streaming model
+            mask (mindspore.tensor): a mask tensor for streaming model
     """
-    chunk_start_idx = ms.Tensor(chunk_start_idx).long()
-    start_pad = mint.nn.functional.pad(
+    chunk_start_idx = mindspore.Tensor(chunk_start_idx).long()
+    start_pad = mindspore.mint.nn.functional.pad(
         chunk_start_idx, (1, 0)
     )  # append 0 to the beginning, so it becomes [0, 0, 18, 36, 48]
-    end_pad = mint.nn.functional.pad(
+    end_pad = mindspore.mint.nn.functional.pad(
         chunk_start_idx, (0, 1), value=x_len
     )  # append x_len to the end, so it becomes [0,18,36,48, x_len]
-    seq_range = mint.arange(0, x_len).unsqueeze(-1)
+    seq_range = mindspore.mint.arange(0, x_len).unsqueeze(-1)
     idx = ((seq_range < end_pad) & (seq_range >= start_pad)).nonzero()[:, 1]
-    seq_range_expand = mint.arange(0, x_len).unsqueeze(0).expand(x_len, -1)
+    seq_range_expand = mindspore.mint.arange(0, x_len).unsqueeze(0).expand(x_len, -1)
     idx_left = idx - left_window
     idx_left[idx_left < 0] = 0
     boundary_left = start_pad[idx_left]
@@ -388,7 +991,7 @@ def __init__(self, config: Phi4MultimodalAudioConfig):
         self.encoder_embedding = Phi4MultimodalAudioMeanVarianceNormLayer(config)
         self.embed = Phi4MultimodalAudioNemoConvSubsampling(config)
         self.relative_attention_bias_layer = Phi4MultimodalAudioRelativeAttentionBias(config)
-        self.encoders = nn.CellList(
+        self.encoders = mindspore.nn.CellList(
             [Phi4MultimodalAudioConformerEncoderLayer(config) for _ in range(config.num_blocks)]
         )
         self.gradient_checkpointing = False
@@ -416,7 +1019,7 @@ def _streaming_mask(self, seq_len, batch_size, chunk_size, left_chunk):
         )
         return enc_streaming_mask
 
-    def construct_embeddings(self, hidden_states, masks):
+    def forward_embeddings(self, hidden_states, masks):
         """Forwarding the inputs through the top embedding layers"""
         seq_len = math.ceil(hidden_states.shape[1] / self.config.time_reduction)
         if seq_len <= 0:
@@ -446,17 +1049,23 @@ def calculate_hs_mask(self, hidden_states, mask):
         enc_streaming_mask = self._streaming_mask(
             max_audio_length, batch_size, self.config.chunk_size, self.config.left_chunk
         )
+        enc_streaming_mask = enc_streaming_mask
         if mask is None:
             return enc_streaming_mask
 
         feature_lens = mask.sum(1)
         padding_length = feature_lens
-        pad_mask = mint.arange(0, max_audio_length).expand(padding_length.size(0), -1) < padding_length.unsqueeze(1)
+        pad_mask = mindspore.mint.arange(
+            0,
+            max_audio_length,
+        ).expand(
+            padding_length.shape[0], -1
+        ) < padding_length.unsqueeze(1)
         pad_mask = pad_mask.unsqueeze(1)
         pad_mask = pad_mask & enc_streaming_mask
         return pad_mask
 
-    def construct(self, hidden_states: ms.Tensor, mask: Optional[ms.Tensor]):
+    def construct(self, hidden_states: mindspore.Tensor, mask: Optional[mindspore.Tensor]):
         hidden_states = self.encoder_embedding(hidden_states)
         hidden_states, hs_mask, mask = self.forward_embeddings(hidden_states, mask)
 
@@ -472,7 +1081,9 @@ def construct(self, hidden_states: ms.Tensor, mask: Optional[ms.Tensor]):
             else:
                 chunk_pad_size = 0
             if chunk_pad_size > 0:
-                hidden_states_pad = F.pad(hidden_states, (0, 0, 0, chunk_pad_size), "constant", 0)
+                hidden_states_pad = mindspore.mint.nn.functional.pad(
+                    hidden_states, (0, 0, 0, chunk_pad_size), "constant", 0
+                )
                 hidden_states = hidden_states_pad
 
             hidden_states = unfold_tensor(hidden_states, max_seq_len)
@@ -480,7 +1091,7 @@ def construct(self, hidden_states: ms.Tensor, mask: Optional[ms.Tensor]):
             if mask is not None:
                 # revise hs_mask here because the previous calculated hs_mask did not consider extra pad
                 subsampled_pad_mask = mask.squeeze(1)  # [bz, subsampled_unmask_seq_len]
-                extra_padded_subsamlped_pad_mask = F.pad(
+                extra_padded_subsamlped_pad_mask = mindspore.mint.nn.functional.pad(
                     subsampled_pad_mask, (0, chunk_pad_size), "constant", False
                 )  # extra padding to the pad mask
                 extra_padded_subsamlped_pad_mask = extra_padded_subsamlped_pad_mask.unsqueeze(-1).float()
@@ -508,34 +1119,36 @@ def construct(self, hidden_states: ms.Tensor, mask: Optional[ms.Tensor]):
         return hidden_states
 
 
-class Phi4MultimodalAudioEmbedding(nn.Cell):
+class Phi4MultimodalAudioEmbedding(mindspore.nn.Cell):
     def __init__(self, config: Phi4MultimodalConfig):
         super().__init__()
         self.config = config
         self.layer_idx = config.audio_config.feature_layer
 
-        self.drop = nn.Dropout(config.embd_pdrop)
+        self.drop = mindspore.mint.nn.Dropout(config.embd_pdrop)
         self.encoder = Phi4MultimodalAudioModel._from_config(config.audio_config)
-        self.up_proj_for_speech = mint.nn.Linear(
+        self.up_proj_for_speech = mindspore.mint.nn.Linear(
             config.audio_config.hidden_size * config.audio_config.downsample_rate, config.hidden_size
         )
-        self.down_proj_for_speech = mint.nn.Linear(config.hidden_size, config.hidden_size)
-        self.up_proj_for_vision_speech = mint.nn.Linear(
+        self.down_proj_for_speech = mindspore.mint.nn.Linear(config.hidden_size, config.hidden_size)
+        self.up_proj_for_vision_speech = mindspore.mint.nn.Linear(
             config.audio_config.hidden_size * config.audio_config.downsample_rate, config.hidden_size
         )
-        self.down_proj_for_vision_speech = mint.nn.Linear(config.hidden_size, config.hidden_size)
+        self.down_proj_for_vision_speech = mindspore.mint.nn.Linear(config.hidden_size, config.hidden_size)
 
     def construct(
         self,
-        input_ids: ms.Tensor,
-        inputs_embeds: ms.Tensor,
-        audio_input_features: ms.Tensor,
+        input_ids: mindspore.Tensor,
+        inputs_embeds: mindspore.Tensor,
+        audio_input_features: mindspore.Tensor,
         audio_embed_sizes=None,
         audio_attention_mask=None,
         audio_projection_mode="speech",
-    ) -> ms.Tensor:
-        with ms._no_grad():
-            positions_tuple = mint.nonzero(input_ids == self.config.audio_config.audio_token_id, as_tuple=True)
+    ) -> mindspore.Tensor:
+        with mindspore._no_grad():
+            positions_tuple = mindspore.mint.nonzero(
+                input_ids == self.config.audio_config.audio_token_id, as_tuple=True
+            )
 
         up_proj = self.up_proj_for_speech if audio_projection_mode == "speech" else self.up_proj_for_vision_speech
         down_proj = self.down_proj_for_speech if audio_projection_mode == "speech" else self.down_proj_for_vision_speech
@@ -546,16 +1159,664 @@ def construct(
 
         audio_encoder_hidden_states = self.encoder(audio_input_features, audio_attention_mask)
         audio_encoder_hidden_states = up_proj(audio_encoder_hidden_states)
-        audio_encoder_hidden_states = mint.nn.functional.gelu(audio_encoder_hidden_states)
+        audio_encoder_hidden_states = mindspore.mint.nn.functional.gelu(audio_encoder_hidden_states)
         audio_embeds = down_proj(audio_encoder_hidden_states)
 
-        merged_audio_embeds = mint.cat(
+        merged_audio_embeds = mindspore.mint.cat(
             [audio_embeds[i, : audio_embed_sizes[i], :] for i in range(len(audio_embed_sizes))], dim=0
         )
-        merged_audio_embeds = merged_audio_embeds.to(dtype=inputs_embeds.dtype)
-
+        merged_audio_embeds = merged_audio_embeds.to(
+            dtype=inputs_embeds.dtype,
+        )
+        # Temporarily disable autocast to avoid issue on bf16 tensors
+        # Ref: https://github.com/pytorch/pytorch/issues/132715
         audio_embeds = inputs_embeds.index_put(indices=positions_tuple, values=merged_audio_embeds, accumulate=False)
 
         audio_embeds = self.drop(audio_embeds)
 
         return audio_embeds
+
+
+class Phi4MultimodalRMSNorm(mindspore.nn.Cell):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        Phi4MultimodalRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = mindspore.Parameter(mindspore.mint.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def construct(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(mindspore.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * mindspore.mint.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
+
+
+class Phi4MultimodalMLP(mindspore.nn.Cell):
+    def __init__(self, config):
+        super().__init__()
+
+        self.config = config
+        self.gate_up_proj = mindspore.mint.nn.Linear(config.hidden_size, 2 * config.intermediate_size, bias=False)
+        self.down_proj = mindspore.mint.nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
+        self.activation_fn = ACT2FN[config.hidden_act]
+
+    def construct(self, hidden_states: mindspore.Tensor) -> mindspore.Tensor:
+        up_states = self.gate_up_proj(hidden_states)
+
+        gate, up_states = up_states.chunk(2, dim=-1)
+        up_states = up_states * self.activation_fn(gate)
+
+        return self.down_proj(up_states)
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return mindspore.mint.cat((-x2, x1), dim=-1)
+
+
+def repeat_kv(hidden_states: mindspore.Tensor, n_rep: int) -> mindspore.Tensor:
+    """
+    This is the equivalent of ms.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def eager_attention_forward(
+    module: mindspore.nn.Cell,
+    query: mindspore.Tensor,
+    key: mindspore.Tensor,
+    value: mindspore.Tensor,
+    attention_mask: Optional[mindspore.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = mindspore.mint.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = mindspore.mint.nn.functional.softmax(attn_weights, dim=-1, dtype=mindspore.float32).to(query.dtype)
+    attn_weights = mindspore.mint.nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = mindspore.mint.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`mindspore.tensor`): The query tensor.
+        k (`mindspore.tensor`): The key tensor.
+        cos (`mindspore.tensor`): The cosine part of the rotary embedding.
+        sin (`mindspore.tensor`): The sine part of the rotary embedding.
+        position_ids (`mindspore.tensor`, *optional*):
+            Deprecated and unused.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(mindspore.tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+
+    rotary_dim = cos.shape[-1]
+    q_rot, q_pass = q[..., :rotary_dim], q[..., rotary_dim:]
+    k_rot, k_pass = k[..., :rotary_dim], k[..., rotary_dim:]
+
+    q_embed = mindspore.mint.cat([(q_rot * cos) + (rotate_half(q_rot) * sin), q_pass], dim=-1)
+    k_embed = mindspore.mint.cat([(k_rot * cos) + (rotate_half(k_rot) * sin), k_pass], dim=-1)
+    return q_embed, k_embed
+
+
+class Phi4MultimodalAttention(mindspore.nn.Cell):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: Phi4MultimodalConfig, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = True
+
+        op_size = config.num_attention_heads * self.head_dim + 2 * (config.num_key_value_heads * self.head_dim)
+        self.o_proj = mindspore.mint.nn.Linear(
+            config.num_attention_heads * self.head_dim, config.hidden_size, bias=False
+        )
+        self.qkv_proj = mindspore.mint.nn.Linear(config.hidden_size, op_size, bias=False)
+
+    def construct(
+        self,
+        hidden_states: mindspore.Tensor,
+        position_embeddings: tuple[mindspore.Tensor, mindspore.Tensor],
+        attention_mask: Optional[mindspore.Tensor],
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[mindspore.Tensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[mindspore.Tensor, Optional[mindspore.Tensor], Optional[tuple[mindspore.Tensor]]]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        qkv = self.qkv_proj(hidden_states)
+        query_pos = self.config.num_attention_heads * self.head_dim
+        query_states = qkv[..., :query_pos]
+        key_states = qkv[..., query_pos : query_pos + self.num_key_value_heads * self.head_dim]
+        value_states = qkv[..., query_pos + self.num_key_value_heads * self.head_dim :]
+
+        query_states = query_states.view(hidden_shape).transpose(1, 2)
+        key_states = key_states.view(hidden_shape).transpose(1, 2)
+        value_states = value_states.view(hidden_shape).transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        if past_key_values is not None:
+            # sin and cos are specific to RoPE models; cache_position needed for the static cache
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            sliding_window=getattr(self.config, "sliding_window", None),
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class Phi4MultimodalDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: Phi4MultimodalConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = Phi4MultimodalAttention(config=config, layer_idx=layer_idx)
+        self.mlp = Phi4MultimodalMLP(config)
+        self.input_layernorm = Phi4MultimodalRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = Phi4MultimodalRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.config = config
+        self.resid_attn_dropout = mindspore.mint.nn.Dropout(config.resid_pdrop)
+        self.resid_mlp_dropout = mindspore.mint.nn.Dropout(config.resid_pdrop)
+
+    def construct(
+        self,
+        hidden_states: mindspore.Tensor,
+        attention_mask: Optional[mindspore.Tensor] = None,
+        position_ids: Optional[mindspore.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[mindspore.Tensor] = None,
+        position_embeddings: Optional[
+            tuple[mindspore.Tensor, mindspore.Tensor]
+        ] = None,  # necessary, but kept here for BC
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[mindspore.Tensor, Optional[tuple[mindspore.Tensor, mindspore.Tensor]]]:
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+
+        hidden_states, self_attn_weights = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = residual + self.resid_attn_dropout(hidden_states)  # main diff with Llama
+
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + self.resid_mlp_dropout(hidden_states)  # main diff with Llama
+        return hidden_states
+
+
+class Phi4MultimodalFeatureEmbedding(mindspore.nn.Cell):
+    """Image-audio embedding."""
+
+    def __init__(self, config: Phi4MultimodalConfig) -> None:
+        super().__init__()
+        self.config = config
+        self.image_token_id = config.vision_config.image_token_id
+        self.audio_token_id = config.audio_config.audio_token_id
+        self.image_embed = Phi4MultimodalImageEmbedding(config)
+        self.audio_embed = Phi4MultimodalAudioEmbedding(config)
+
+    def construct(
+        self,
+        input_ids: mindspore.Tensor,
+        inputs_embeds: mindspore.Tensor,
+        image_pixel_values: Optional[mindspore.Tensor] = None,
+        audio_input_features: Optional[mindspore.Tensor] = None,
+        image_sizes=None,
+        image_attention_mask=None,
+        audio_embed_sizes=None,
+        audio_attention_mask=None,
+    ) -> mindspore.Tensor:
+        with mindspore._no_grad():
+            image_position_mask = (input_ids == self.config.vision_config.image_token_id).unsqueeze(-1)
+            non_image_position_mask = ~image_position_mask
+
+        image_embeds = None
+        audio_embeds = None
+        if image_pixel_values is not None and (input_ids == self.image_token_id).any():
+            image_embeds = self.image_embed(
+                input_ids,
+                inputs_embeds,
+                image_pixel_values=image_pixel_values,
+                image_sizes=image_sizes,
+                image_attention_mask=image_attention_mask,
+            )
+        if audio_input_features is not None and (input_ids == self.audio_token_id).any():
+            audio_projection_mode = "vision" if image_pixel_values is not None else "speech"
+            audio_embeds = self.audio_embed(
+                input_ids,
+                inputs_embeds,
+                audio_input_features=audio_input_features,
+                audio_embed_sizes=audio_embed_sizes,
+                audio_attention_mask=audio_attention_mask,
+                audio_projection_mode=audio_projection_mode,
+            )
+
+        # merge image and audio
+        if image_embeds is not None and audio_embeds is not None:
+            inputs_embeds = image_embeds * image_position_mask + audio_embeds * non_image_position_mask
+        elif image_embeds is not None:
+            inputs_embeds = image_embeds
+        elif audio_embeds is not None:
+            inputs_embeds = audio_embeds
+
+        return inputs_embeds
+
+
+class Phi4MultimodalRotaryEmbedding(mindspore.nn.Cell):
+    inv_freq: mindspore.Tensor  # fix linting for `register_buffer`
+
+    def __init__(self, config: Phi4MultimodalConfig):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and isinstance(config.rope_scaling, dict):
+            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
+        else:
+            self.rope_type = "default"
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.original_inv_freq = self.inv_freq
+
+    @mindspore._no_grad()
+    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
+    def construct(self, x, position_ids):
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        # Force float32
+        freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+        emb = mindspore.mint.cat((freqs, freqs), dim=-1)
+        cos = emb.cos() * self.attention_scaling
+        sin = emb.sin() * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+class Phi4MultimodalPreTrainedModel(PreTrainedModel):
+    config: Phi4MultimodalConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["Phi4MultimodalDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+
+    _can_compile_fullgraph = True
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "hidden_states": Phi4MultimodalDecoderLayer,
+        "attentions": Phi4MultimodalAttention,
+    }
+    _version = "0.0.5"
+
+    def _init_weights(self, module):
+        super()._init_weights(module)
+        if isinstance(module, Phi4MultimodalImageEmbedding):
+            module.global_img_feature_extensor.data.zero_()
+            module.sub_img_feature_extensor.data.zero_()
+
+
+class Phi4MultimodalModel(Phi4MultimodalPreTrainedModel):
+    def __init__(self, config: Phi4MultimodalConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = mindspore.mint.nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+
+        self.layers = mindspore.nn.CellList(
+            [Phi4MultimodalDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = Phi4MultimodalRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = Phi4MultimodalRotaryEmbedding(config=config)
+
+        self.gradient_checkpointing = False
+        self.embed_dropout = mindspore.mint.nn.Dropout(config.embd_pdrop)
+
+        self.embed_tokens_extend = Phi4MultimodalFeatureEmbedding(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @check_model_inputs
+    def construct(
+        self,
+        input_ids: Optional[mindspore.Tensor] = None,
+        attention_mask: Optional[mindspore.Tensor] = None,
+        position_ids: Optional[mindspore.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[mindspore.Tensor] = None,
+        image_pixel_values: Optional[mindspore.Tensor] = None,
+        image_sizes: Optional[mindspore.Tensor] = None,
+        image_attention_mask=None,
+        audio_input_features: Optional[mindspore.Tensor] = None,
+        audio_embed_sizes=None,
+        audio_attention_mask=None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        cache_position: Optional[mindspore.Tensor] = None,
+        **kwargs,
+    ) -> BaseModelOutputWithPast:
+        r"""
+        image_pixel_values (`mindspore.tensor`, *optional*):
+            If the input contains images, these correspond to the pixel values after transformations (as returned by
+            the Processor)
+        image_sizes (`mindspore.tensor`, *optional*):
+            If the input contains images, these correspond to size of each image.
+        image_attention_mask (`mindspore.tensor`, *optional*):
+            Attention mask for the images.
+        audio_input_features (`mindspore.tensor`, *optional*):
+            If the input contains audio samples, these correspond to the values after transformation (as returned by
+            the Processor).
+        audio_embed_sizes (`mindspore.tensor`, *optional*):
+            Size of the audio inputs.
+        audio_attention_mask (`mindspore.tensor, *optional*):
+            Attention mask for the audio inputs.
+        """
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache(config=self.config)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+            inputs_embeds = self.embed_tokens_extend(
+                input_ids,
+                inputs_embeds,
+                image_pixel_values=image_pixel_values,
+                audio_input_features=audio_input_features,
+                image_sizes=image_sizes,
+                image_attention_mask=image_attention_mask,
+                audio_embed_sizes=audio_embed_sizes,
+                audio_attention_mask=audio_attention_mask,
+            )
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = mindspore.mint.arange(
+                past_seen_tokens,
+                past_seen_tokens + inputs_embeds.shape[1],
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        mask_function = create_causal_mask if self.config.sliding_window is None else create_sliding_window_causal_mask
+        causal_mask = mask_function(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+
+        # create position embeddings to be shared across the decoder layers
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+        for decoder_layer in self.layers:
+            hidden_states = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_ids=position_ids,
+                past_key_values=past_key_values,
+                use_cache=use_cache,
+                cache_position=cache_position,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+
+        hidden_states = self.norm(hidden_states)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values if use_cache else None,
+        )
+
+
+class Phi4MultimodalForCausalLM(Phi4MultimodalPreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+    _tp_plan = {"lm_head": "colwise_rep"}
+    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = Phi4MultimodalModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = mindspore.mint.nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @can_return_tuple
+    def construct(
+        self,
+        input_ids: Optional[mindspore.Tensor] = None,
+        attention_mask: Optional[mindspore.Tensor] = None,
+        position_ids: Optional[mindspore.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[mindspore.Tensor] = None,
+        image_pixel_values: Optional[mindspore.Tensor] = None,
+        image_sizes: Optional[mindspore.Tensor] = None,
+        image_attention_mask=None,
+        audio_input_features: Optional[mindspore.Tensor] = None,
+        audio_embed_sizes=None,
+        audio_attention_mask=None,
+        labels: Optional[mindspore.Tensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        cache_position: Optional[mindspore.Tensor] = None,
+        logits_to_keep: Union[int, mindspore.Tensor] = 0,
+        **kwargs,
+    ) -> CausalLMOutputWithPast:
+        r"""
+        image_pixel_values (`mindspore.tensor`, *optional*):
+            If the input contains images, these correspond to the pixel values after transformations (as returned by
+            the Processor)
+        image_sizes (`mindspore.tensor`, *optional*):
+            If the input contains images, these correspond to size of each image.
+        image_attention_mask (`mindspore.tensor`, *optional*):
+            Attention mask for the images.
+        audio_input_features (`mindspore.tensor`, *optional*):
+            If the input contains audio samples, these correspond to the values after transformation (as returned by
+            the Processor).
+        audio_embed_sizes (`mindspore.tensor`, *optional*):
+            Size of the audio inputs.
+        audio_attention_mask (`mindspore.tensor, *optional*):
+            Attention mask for the audio inputs.
+        labels (`mindspore.tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Example:
+        ```python
+        >>> from transformers import AutoTokenizer
+        >>> from mindone.transformers import Phi4MultimodalForCausalLM
+        >>> import mindspore as ms
+        >>> model = Phi4MultimodalForCausalLM.from_pretrained("TBA")
+        >>> tokenizer = AutoTokenizer.from_pretrained("TBA")
+        >>> prompt = "This is an example script ."
+        >>> inputs = tokenizer(prompt, return_tensors="np")
+        >>> # Generate
+        >>> generate_ids = model.generate(ms.tensor(inputs.input_ids), max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        'This is an example script .\n Certainly! Below is a sample script that demonstrates a simple task, such as calculating the sum'
+        ```"""
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs: BaseModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            image_pixel_values=image_pixel_values,
+            image_sizes=image_sizes,
+            image_attention_mask=image_attention_mask,
+            audio_input_features=audio_input_features,
+            audio_embed_sizes=audio_embed_sizes,
+            audio_attention_mask=audio_attention_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits, labels, self.vocab_size)
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        inputs_embeds=None,
+        image_pixel_values=None,
+        image_sizes=None,
+        image_attention_mask=None,
+        audio_input_features=None,
+        audio_embed_sizes=None,
+        audio_attention_mask=None,
+        cache_position=None,
+        position_ids=None,
+        use_cache=True,
+        logits_to_keep=0,
+        **kwargs,
+    ):
+        # Overwritten -- this model may need to switch between short and long rope, invalidating the cache in the
+        # process
+
+        # When the first time input length reached long and short factor switching point, enforce re-compute cache
+        # It will cause downside of slower at this single token position, however, better than current failure.
+        if (
+            past_key_values
+            and self.config.rope_scaling
+            and input_ids.shape[1] >= self.config.original_max_position_embeddings + 1
+        ):
+            past_length = cache_position[0]
+            if past_length <= self.config.original_max_position_embeddings:
+                past_key_values = None
+
+        model_inputs = super().prepare_inputs_for_generation(
+            input_ids=input_ids,
+            past_key_values=past_key_values,
+            attention_mask=attention_mask,
+            inputs_embeds=inputs_embeds,
+            image_pixel_values=image_pixel_values,
+            image_sizes=image_sizes,
+            image_attention_mask=image_attention_mask,
+            audio_input_features=audio_input_features,
+            audio_embed_sizes=audio_embed_sizes,
+            audio_attention_mask=audio_attention_mask,
+            cache_position=cache_position,
+            position_ids=position_ids,
+            use_cache=use_cache,
+            logits_to_keep=logits_to_keep,
+            **kwargs,
+        )
+        return model_inputs
+
+
+__all__ = [
+    "Phi4MultimodalAudioPreTrainedModel",
+    "Phi4MultimodalAudioModel",
+    "Phi4MultimodalVisionPreTrainedModel",
+    "Phi4MultimodalVisionModel",
+    "Phi4MultimodalPreTrainedModel",
+    "Phi4MultimodalModel",
+    "Phi4MultimodalForCausalLM",
+]
diff --git a/tests/transformers_tests/models/phi4_multimodal/__init__.py b/tests/transformers_tests/models/phi4_multimodal/__init__.py
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/tests/transformers_tests/models/phi4_multimodal/test_modeling_phi4_multimodal.py b/tests/transformers_tests/models/phi4_multimodal/test_modeling_phi4_multimodal.py
new file mode 100644
index 0000000000..288cd78bfa
--- /dev/null
+++ b/tests/transformers_tests/models/phi4_multimodal/test_modeling_phi4_multimodal.py
@@ -0,0 +1,251 @@
+"""Adapted from https://github.com/huggingface/transformers/tree/main/tests//models/phi4_multimodal/test_modeling_phi4_multimodal.py."""
+
+# This module contains test cases that are defined in the `.test_cases.py` file, structured as lists or tuples like
+#     [name, pt_module, ms_module, init_args, init_kwargs, inputs_args, inputs_kwargs, outputs_map].
+#
+# Each defined case corresponds to a pair consisting of PyTorch and MindSpore modules, including their respective
+# initialization parameters and inputs for the forward. The testing framework adopted here is designed to generically
+# parse these parameters to assess and compare the precision of forward outcomes between the two frameworks.
+#
+# In cases where models have unique initialization procedures or require testing with specialized output formats,
+# it is necessary to develop distinct, dedicated test cases.
+
+import inspect
+import logging
+
+import numpy as np
+import pytest
+import torch
+from transformers import Phi4MultimodalAudioConfig, Phi4MultimodalConfig, Phi4MultimodalVisionConfig
+
+import mindspore as ms
+
+from tests.modeling_test_utils import (
+    MS_DTYPE_MAPPING,
+    PT_DTYPE_MAPPING,
+    compute_diffs,
+    generalized_parse_args,
+    get_modules,
+)
+from tests.transformers_tests.models.modeling_common import floats_numpy, ids_numpy
+
+DTYPE_AND_THRESHOLDS = {"fp32": 5e-4, "fp16": 5e-3, "bf16": 5e-3}
+MODES = [1]
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+
+class Phi4MultimodalModelTester:
+    def __init__(
+        self,
+        batch_size=2,
+        seq_length=12,
+        image_seq_length=275,
+        audio_seq_length=8,
+        is_training=True,
+        num_hidden_layers=2,
+        vocab_size=49,
+        hidden_size=32,
+        intermediate_size=64,
+        num_attention_heads=8,
+        num_key_value_heads=4,
+        bos_token_id=0,
+        eos_token_id=0,
+        pad_token_id=0,
+        image_token_id=1,
+        audio_token_id=2,
+        image_size=16,
+        audio_size=12,
+        audio_config=Phi4MultimodalAudioConfig(
+            num_blocks=2,
+            hidden_size=32,
+            num_attention_heads=8,
+            intermediate_size=48,
+            depthwise_separable_out_channel=128,
+            nemo_conv_channels=128,
+            initializer_range=1e-5,
+        ),
+        vision_config=Phi4MultimodalVisionConfig(
+            num_hidden_layers=2,
+            hidden_size=32,
+            intermediate_size=64,
+            num_attention_heads=8,
+            crop_size=16,
+            initializer_range=1e-5,
+        ),
+    ):
+        self.num_hidden_layers = num_hidden_layers
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_attention_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        self.bos_token_id = bos_token_id
+        self.pad_token_id = pad_token_id
+        self.eos_token_id = eos_token_id
+        self.image_token_id = image_token_id
+        self.audio_token_id = audio_token_id
+        self.audio_config = audio_config
+        self.vision_config = vision_config
+
+        self.is_training = is_training
+        self.batch_size = batch_size
+        self.seq_length = seq_length + image_seq_length + audio_seq_length
+        self.image_seq_length = image_seq_length
+        self.audio_seq_length = audio_seq_length
+        self.image_size = image_size
+        self.audio_size = audio_size
+        self.num_channels = 3
+
+    def get_config(self):
+        return Phi4MultimodalConfig(
+            num_hidden_layers=self.num_hidden_layers,
+            vocab_size=self.vocab_size,
+            hidden_size=self.hidden_size,
+            intermediate_size=self.intermediate_size,
+            num_attention_heads=self.num_attention_heads,
+            num_key_value_heads=self.num_key_value_heads,
+            bos_token_id=self.bos_token_id,
+            eos_token_id=self.eos_token_id,
+            pad_token_id=self.pad_token_id,
+            vision_config=self.vision_config,
+            audio_config=self.audio_config,
+        )
+
+    # Copied from tests.models.mistral.test_modeling_mistral.MistralModelTester.prepare_config_and_inputs
+    def prepare_config_and_inputs(self):
+        input_ids = ids_numpy([self.batch_size, self.seq_length], self.vocab_size)
+
+        # The shapes corresponds to the inputs for image of size 16x16
+        image_pixel_values = floats_numpy([self.batch_size, 2, self.num_channels, self.image_size, self.image_size])
+        image_attention_mask = np.ones((self.batch_size, 2, 1, 1))
+        image_sizes = ms.tensor([[self.image_size, self.image_size]] * self.batch_size, dtype=ms.int64)
+
+        # Feature sizes returned by an audio of size 10000
+        audio_input_features = floats_numpy([self.batch_size, 61, 80])
+        audio_embed_sizes = ms.tensor([self.audio_seq_length] * self.batch_size, dtype=ms.int64)
+
+        input_ids[input_ids == self.pad_token_id] = self.pad_token_id + 1  # random value but not pad token
+        input_ids[-1, 0] = self.pad_token_id  # mask the last text token
+        input_ids[:, -self.image_seq_length - self.audio_seq_length : -self.audio_seq_length] = self.image_token_id
+        input_ids[:, -self.audio_seq_length :] = self.audio_token_id
+
+        attention_mask = np.ones_like(input_ids)
+        attention_mask[-1, 0] = 0  # mask the last text token
+        config = self.get_config()
+
+        return (
+            config,
+            input_ids,
+            attention_mask,
+            image_pixel_values,
+            image_attention_mask,
+            image_sizes,
+            audio_input_features,
+            audio_embed_sizes,
+        )
+
+
+model_tester = Phi4MultimodalModelTester()
+(
+    config,
+    input_ids,
+    attention_mask,
+    image_pixel_values,
+    image_attention_mask,
+    image_sizes,
+    audio_input_features,
+    audio_embed_sizes,
+) = model_tester.prepare_config_and_inputs()
+
+
+PHI4_CASES = [
+    [
+        "Phi4MultimodalModel",
+        "transformers.Phi4MultimodalModel",
+        "mindone.transformers.Phi4MultimodalModel",
+        (config,),
+        {},
+        (input_ids,),
+        {
+            "attention_mask": attention_mask,
+        },
+        {
+            "last_hidden_state": 0,
+        },
+    ],
+]
+
+
+# transformers need >= 4.41.2
+@pytest.mark.parametrize(
+    "name,pt_module,ms_module,init_args,init_kwargs,inputs_args,inputs_kwargs,outputs_map,dtype,mode",
+    [
+        case
+        + [
+            dtype,
+        ]
+        + [
+            mode,
+        ]
+        for case in PHI4_CASES
+        for dtype in DTYPE_AND_THRESHOLDS.keys()
+        for mode in MODES
+    ],
+)
+def test_named_modules(
+    name,
+    pt_module,
+    ms_module,
+    init_args,
+    init_kwargs,
+    inputs_args,
+    inputs_kwargs,
+    outputs_map,
+    dtype,
+    mode,
+):
+    ms.set_context(mode=mode)
+
+    (
+        pt_model,
+        ms_model,
+        pt_dtype,
+        ms_dtype,
+    ) = get_modules(pt_module, ms_module, dtype, *init_args, **init_kwargs)
+    pt_inputs_args, pt_inputs_kwargs, ms_inputs_args, ms_inputs_kwargs = generalized_parse_args(
+        pt_dtype, ms_dtype, *inputs_args, **inputs_kwargs
+    )
+
+    # set `hidden_dtype` if requiring, for some modules always compute in float
+    # precision and require specific `hidden_dtype` to cast before return
+    if "hidden_dtype" in inspect.signature(pt_model.forward).parameters:
+        pt_inputs_kwargs.update({"hidden_dtype": PT_DTYPE_MAPPING[pt_dtype]})
+        ms_inputs_kwargs.update({"hidden_dtype": MS_DTYPE_MAPPING[ms_dtype]})
+    with torch.no_grad():
+        pt_outputs = pt_model(*pt_inputs_args, **pt_inputs_kwargs)
+    ms_outputs = ms_model(*ms_inputs_args, **ms_inputs_kwargs)
+    # logger.info(f"ms:{ms_outputs}")
+    # logger.info(f"pt:{pt_outputs}" )
+    if outputs_map:
+        pt_outputs_n = []
+        ms_outputs_n = []
+        for pt_key, ms_idx in outputs_map.items():
+            pt_output = getattr(pt_outputs, pt_key)
+            ms_output = ms_outputs[ms_idx]
+            if isinstance(pt_output, (list, tuple)):
+                pt_outputs_n += list(pt_output)
+                ms_outputs_n += list(ms_output)
+            else:
+                pt_outputs_n.append(pt_output)
+                ms_outputs_n.append(ms_output)
+        diffs = compute_diffs(pt_outputs_n, ms_outputs_n)
+    else:
+        diffs = compute_diffs(pt_outputs, ms_outputs)
+    logger.info(f"Differences: {diffs}")
+    THRESHOLD = DTYPE_AND_THRESHOLDS[ms_dtype]
+    assert (np.array(diffs) < THRESHOLD).all(), (
+        f"ms_dtype: {ms_dtype}, pt_type: {pt_dtype}, "
+        f"Outputs({np.array(diffs).tolist()}) has diff bigger than {THRESHOLD}"
+    )