[Refactor] [7/N] to simplify the vLLM lora serving architecture (#32251)

Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>

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"

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"]
 

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

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__)

vllm/entrypoints/serve/lora/protocol.py

+# 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)