diff --git a/tests/models/test_deepseek_v4_mega_moe.py b/tests/models/test_deepseek_v4_mega_moe.py
index 24a8701e702..3d4657bba13 100644
--- a/tests/models/test_deepseek_v4_mega_moe.py
+++ b/tests/models/test_deepseek_v4_mega_moe.py
@@ -6,7 +6,7 @@ from types import SimpleNamespace
 import pytest
 import torch
 
-from vllm.models.deepseek_v4.nvidia.deepseek_v4 import (
+from vllm.models.deepseek_v4.nvidia.model import (
     DeepseekV4MegaMoEExperts,
     _stage_deepseek_v4_mega_moe_inputs,
     make_deepseek_v4_expert_params_mapping,
diff --git a/vllm/models/deepseek_v4/__init__.py b/vllm/models/deepseek_v4/__init__.py
index a8c0fc22240..abaa794f98c 100644
--- a/vllm/models/deepseek_v4/__init__.py
+++ b/vllm/models/deepseek_v4/__init__.py
@@ -17,11 +17,11 @@ from .quant_config import DeepseekV4FP8Config
 # default that mypy sees; the ROCm branch overrides it at runtime and is
 # kept type-compatible via ``# type: ignore[assignment]``.
 if TYPE_CHECKING or not current_platform.is_rocm():
-    from .nvidia.deepseek_v4 import DeepseekV4ForCausalLM
-    from .nvidia.deepseek_v4_mtp import DeepSeekV4MTP
+    from .nvidia.model import DeepseekV4ForCausalLM
+    from .nvidia.mtp import DeepSeekV4MTP
 else:
-    from .amd.deepseek_v4 import DeepseekV4ForCausalLM  # type: ignore[assignment]
-    from .amd.deepseek_v4_mtp import DeepSeekV4MTP  # type: ignore[assignment]
+    from .amd.model import DeepseekV4ForCausalLM  # type: ignore[assignment]
+    from .amd.mtp import DeepSeekV4MTP  # type: ignore[assignment]
 
 __all__ = [
     "DeepSeekV4MTP",
diff --git a/vllm/models/deepseek_v4/amd/deepseek_v4.py b/vllm/models/deepseek_v4/amd/deepseek_v4.py
deleted file mode 120000
index b16a47b66d2..00000000000
--- a/vllm/models/deepseek_v4/amd/deepseek_v4.py
+++ /dev/null
@@ -1 +0,0 @@
-../nvidia/deepseek_v4.py
\ No newline at end of file
diff --git a/vllm/models/deepseek_v4/amd/deepseek_v4_mtp.py b/vllm/models/deepseek_v4/amd/deepseek_v4_mtp.py
deleted file mode 120000
index db212511d8a..00000000000
--- a/vllm/models/deepseek_v4/amd/deepseek_v4_mtp.py
+++ /dev/null
@@ -1 +0,0 @@
-../nvidia/deepseek_v4_mtp.py
\ No newline at end of file
diff --git a/vllm/models/deepseek_v4/amd/model.py b/vllm/models/deepseek_v4/amd/model.py
new file mode 120000
index 00000000000..b09c9c7e35c
--- /dev/null
+++ b/vllm/models/deepseek_v4/amd/model.py
@@ -0,0 +1 @@
+../nvidia/model.py
\ No newline at end of file
diff --git a/vllm/models/deepseek_v4/amd/mtp.py b/vllm/models/deepseek_v4/amd/mtp.py
new file mode 120000
index 00000000000..ac117939dc3
--- /dev/null
+++ b/vllm/models/deepseek_v4/amd/mtp.py
@@ -0,0 +1 @@
+../nvidia/mtp.py
\ No newline at end of file
diff --git a/vllm/models/deepseek_v4/nvidia/deepseek_v4.py b/vllm/models/deepseek_v4/nvidia/model.py
similarity index 100%
rename from vllm/models/deepseek_v4/nvidia/deepseek_v4.py
rename to vllm/models/deepseek_v4/nvidia/model.py
diff --git a/vllm/models/deepseek_v4/nvidia/deepseek_v4_mtp.py b/vllm/models/deepseek_v4/nvidia/mtp.py
similarity index 99%
rename from vllm/models/deepseek_v4/nvidia/deepseek_v4_mtp.py
rename to vllm/models/deepseek_v4/nvidia/mtp.py
index 41060243eb4..071abe2f4a4 100644
--- a/vllm/models/deepseek_v4/nvidia/deepseek_v4_mtp.py
+++ b/vllm/models/deepseek_v4/nvidia/mtp.py
@@ -40,7 +40,7 @@ from vllm.model_executor.models.utils import maybe_prefix
 from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors
 
-from .deepseek_v4 import (
+from .model import (
     DeepseekV4DecoderLayer,
     make_deepseek_v4_expert_params_mapping,
 )