diff --git a/tests/entrypoints/openai/test_serving_models.py b/tests/entrypoints/openai/test_serving_models.py
index e596b32ba..a671611c7 100644
--- a/tests/entrypoints/openai/test_serving_models.py
+++ b/tests/entrypoints/openai/test_serving_models.py
@@ -10,10 +10,12 @@ from vllm.config import ModelConfig
 from vllm.engine.protocol import EngineClient
 from vllm.entrypoints.openai.engine.protocol import (
     ErrorResponse,
+)
+from vllm.entrypoints.openai.serving_models import BaseModelPath, OpenAIServingModels
+from vllm.entrypoints.serve.lora.protocol import (
     LoadLoRAAdapterRequest,
     UnloadLoRAAdapterRequest,
 )
-from vllm.entrypoints.openai.serving_models import BaseModelPath, OpenAIServingModels
 from vllm.lora.request import LoRARequest
 
 MODEL_NAME = "hmellor/tiny-random-LlamaForCausalLM"
diff --git a/vllm/entrypoints/openai/engine/protocol.py b/vllm/entrypoints/openai/engine/protocol.py
index 9434e214f..4ada3f073 100644
--- a/vllm/entrypoints/openai/engine/protocol.py
+++ b/vllm/entrypoints/openai/engine/protocol.py
@@ -1247,16 +1247,6 @@ StreamingResponsesResponse: TypeAlias = (
 )
 
 
-class LoadLoRAAdapterRequest(BaseModel):
-    lora_name: str
-    lora_path: str
-
-
-class UnloadLoRAAdapterRequest(BaseModel):
-    lora_name: str
-    lora_int_id: int | None = Field(default=None)
-
-
 ## Protocols for Audio
 AudioResponseFormat: TypeAlias = Literal["json", "text", "srt", "verbose_json", "vtt"]
 
diff --git a/vllm/entrypoints/openai/serving_models.py b/vllm/entrypoints/openai/serving_models.py
index 4e6d4d5e0..4fed2a827 100644
--- a/vllm/entrypoints/openai/serving_models.py
+++ b/vllm/entrypoints/openai/serving_models.py
@@ -10,10 +10,12 @@ from vllm.engine.protocol import EngineClient
 from vllm.entrypoints.openai.engine.protocol import (
     ErrorInfo,
     ErrorResponse,
-    LoadLoRAAdapterRequest,
     ModelCard,
     ModelList,
     ModelPermission,
+)
+from vllm.entrypoints.serve.lora.protocol import (
+    LoadLoRAAdapterRequest,
     UnloadLoRAAdapterRequest,
 )
 from vllm.logger import init_logger
diff --git a/vllm/entrypoints/serve/lora/api_router.py b/vllm/entrypoints/serve/lora/api_router.py
index 41ec354d7..dd6f692ce 100644
--- a/vllm/entrypoints/serve/lora/api_router.py
+++ b/vllm/entrypoints/serve/lora/api_router.py
@@ -10,10 +10,12 @@ from vllm import envs
 from vllm.entrypoints.openai.api_server import models, validate_json_request
 from vllm.entrypoints.openai.engine.protocol import (
     ErrorResponse,
+)
+from vllm.entrypoints.openai.serving_models import OpenAIServingModels
+from vllm.entrypoints.serve.lora.protocol import (
     LoadLoRAAdapterRequest,
     UnloadLoRAAdapterRequest,
 )
-from vllm.entrypoints.openai.serving_models import OpenAIServingModels
 from vllm.logger import init_logger
 
 logger = init_logger(__name__)
diff --git a/vllm/entrypoints/serve/lora/protocol.py b/vllm/entrypoints/serve/lora/protocol.py
new file mode 100644
index 000000000..e39f35f38
--- /dev/null
+++ b/vllm/entrypoints/serve/lora/protocol.py
@@ -0,0 +1,14 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from pydantic import BaseModel, Field
+
+
+class LoadLoRAAdapterRequest(BaseModel):
+    lora_name: str
+    lora_path: str
+
+
+class UnloadLoRAAdapterRequest(BaseModel):
+    lora_name: str
+    lora_int_id: int | None = Field(default=None)