A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 10, 05:15 UTC
[Refactor] [7/N] to simplify the vLLM lora serving architecture (#32251)
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
Chauncey committed
Jan 13, 2026 at 23:37 UTC
4f02cb2eac20a1281784914acca80cd1fe2bfd15
5 files changed
+23
-13
tests/entrypoints/openai/test_serving_models.py
+3
-1
@@ -10,10 +10,12 @@ from vllm.config import ModelConfig
10
from vllm.engine.protocol import EngineClient
11
from vllm.entrypoints.openai.engine.protocol import (
12
ErrorResponse,
13
+)
14
+from vllm.entrypoints.openai.serving_models import BaseModelPath, OpenAIServingModels
15
+from vllm.entrypoints.serve.lora.protocol import (
16
LoadLoRAAdapterRequest,
17
UnloadLoRAAdapterRequest,
18
)
16
-from vllm.entrypoints.openai.serving_models import BaseModelPath, OpenAIServingModels
19
from vllm.lora.request import LoRARequest
20
21
MODEL_NAME = "hmellor/tiny-random-LlamaForCausalLM"
vllm/entrypoints/openai/engine/protocol.py
-10
@@ -1247,16 +1247,6 @@ StreamingResponsesResponse: TypeAlias = (
1247
)
1248
1249
1250
-class LoadLoRAAdapterRequest(BaseModel):
1251
- lora_name: str
1252
- lora_path: str
1253
-
1254
-
1255
-class UnloadLoRAAdapterRequest(BaseModel):
1256
- lora_name: str
1257
- lora_int_id: int | None = Field(default=None)
1258
-
1259
-
1250
## Protocols for Audio
1251
AudioResponseFormat: TypeAlias = Literal["json", "text", "srt", "verbose_json", "vtt"]
1252
vllm/entrypoints/openai/serving_models.py
+3
-1
@@ -10,10 +10,12 @@ from vllm.engine.protocol import EngineClient
10
from vllm.entrypoints.openai.engine.protocol import (
11
ErrorInfo,
12
ErrorResponse,
13
- LoadLoRAAdapterRequest,
13
ModelCard,
14
ModelList,
15
ModelPermission,
16
+)
17
+from vllm.entrypoints.serve.lora.protocol import (
18
+ LoadLoRAAdapterRequest,
19
UnloadLoRAAdapterRequest,
20
)
21
from vllm.logger import init_logger
vllm/entrypoints/serve/lora/api_router.py
+3
-1
@@ -10,10 +10,12 @@ from vllm import envs
10
from vllm.entrypoints.openai.api_server import models, validate_json_request
11
from vllm.entrypoints.openai.engine.protocol import (
12
ErrorResponse,
13
+)
14
+from vllm.entrypoints.openai.serving_models import OpenAIServingModels
15
+from vllm.entrypoints.serve.lora.protocol import (
16
LoadLoRAAdapterRequest,
17
UnloadLoRAAdapterRequest,
18
)
16
-from vllm.entrypoints.openai.serving_models import OpenAIServingModels
19
from vllm.logger import init_logger
20
21
logger = init_logger(__name__)
vllm/entrypoints/serve/lora/protocol.py
new
+14
@@ -0,0 +1,14 @@
1
+# SPDX-License-Identifier: Apache-2.0
2
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
+
4
+from pydantic import BaseModel, Field
5
+
6
+
7
+class LoadLoRAAdapterRequest(BaseModel):
8
+ lora_name: str
9
+ lora_path: str
10
+
11
+
12
+class UnloadLoRAAdapterRequest(BaseModel):
13
+ lora_name: str
14
+ lora_int_id: int | None = Field(default=None)