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 13, 03:15 UTC
[RL] [V1] Remove unused device argument from reset_kv_cache (#28766)
Signed-off-by: Zhuohan Li <zhuohan123@gmail.com>
Zhuohan Li committed
Nov 14, 2025 at 23:59 UTC
dd6ac1c2bb3d29f8ba612a2f66f350a2c55c7e8b
5 files changed
+9
-17
vllm/engine/protocol.py
+1
-1
@@ -125,7 +125,7 @@ class EngineClient(ABC):
125
...
126
127
@abstractmethod
128
- async def reset_prefix_cache(self, device: Device | None = None) -> None:
128
+ async def reset_prefix_cache(self) -> None:
129
"""Reset the prefix cache"""
130
...
131
vllm/entrypoints/llm.py
+2
-3
@@ -32,7 +32,6 @@ from vllm.config.model import (
32
TokenizerMode,
33
)
34
from vllm.engine.arg_utils import EngineArgs
35
-from vllm.engine.protocol import Device
35
from vllm.entrypoints.chat_utils import (
36
ChatCompletionMessageParam,
37
ChatTemplateContentFormatOption,
@@ -1499,8 +1498,8 @@ class LLM:
1498
def stop_profile(self) -> None:
1499
self.llm_engine.stop_profile()
1500
1502
- def reset_prefix_cache(self, device: Device | None = None) -> None:
1503
- self.llm_engine.reset_prefix_cache(device)
1501
+ def reset_prefix_cache(self) -> None:
1502
+ self.llm_engine.reset_prefix_cache()
1503
1504
def sleep(self, level: int = 1):
1505
"""
vllm/entrypoints/openai/api_server.py
+3
-7
@@ -39,7 +39,7 @@ from typing_extensions import assert_never
39
import vllm.envs as envs
40
from vllm.config import VllmConfig
41
from vllm.engine.arg_utils import AsyncEngineArgs
42
-from vllm.engine.protocol import Device, EngineClient
42
+from vllm.engine.protocol import EngineClient
43
from vllm.entrypoints.anthropic.protocol import (
44
AnthropicError,
45
AnthropicErrorResponse,
@@ -1069,12 +1069,8 @@ if envs.VLLM_SERVER_DEV_MODE:
1069
Reset the prefix cache. Note that we currently do not check if the
1070
prefix cache is successfully reset in the API server.
1071
"""
1072
- device = None
1073
- device_str = raw_request.query_params.get("device")
1074
- if device_str is not None:
1075
- device = Device[device_str.upper()]
1076
- logger.info("Resetting prefix cache with specific %s...", str(device))
1077
- await engine_client(raw_request).reset_prefix_cache(device)
1072
+ logger.info("Resetting prefix cache...")
1073
+ await engine_client(raw_request).reset_prefix_cache()
1074
return Response(status_code=200)
1075
1076
@router.post("/reset_mm_cache")
vllm/v1/engine/async_llm.py
+2
-4
@@ -14,7 +14,7 @@ import torch
14
import vllm.envs as envs
15
from vllm.config import VllmConfig
16
from vllm.engine.arg_utils import AsyncEngineArgs
17
-from vllm.engine.protocol import Device, EngineClient
17
+from vllm.engine.protocol import EngineClient
18
from vllm.entrypoints.utils import _validate_truncation_size
19
from vllm.inputs import PromptType
20
from vllm.logger import init_logger
@@ -672,9 +672,7 @@ class AsyncLLM(EngineClient):
672
self.processor.clear_mm_cache()
673
await self.engine_core.reset_mm_cache_async()
674
675
- async def reset_prefix_cache(self, device: Device | None = None) -> None:
676
- if device == Device.CPU:
677
- raise ValueError("Not supported on CPU.")
675
+ async def reset_prefix_cache(self) -> None:
676
await self.engine_core.reset_prefix_cache_async()
677
678
async def sleep(self, level: int = 1) -> None:
vllm/v1/engine/llm_engine.py
+1
-2
@@ -14,7 +14,6 @@ from vllm.config import ParallelConfig, VllmConfig
14
from vllm.distributed import stateless_destroy_torch_distributed_process_group
15
from vllm.distributed.parallel_state import get_dp_group
16
from vllm.engine.arg_utils import EngineArgs
17
-from vllm.engine.protocol import Device
17
from vllm.inputs import PromptType
18
from vllm.logger import init_logger
19
from vllm.lora.request import LoRARequest
@@ -321,7 +320,7 @@ class LLMEngine:
320
self.processor.clear_mm_cache()
321
self.engine_core.reset_mm_cache()
322
324
- def reset_prefix_cache(self, device: Device | None = None):
323
+ def reset_prefix_cache(self):
324
self.engine_core.reset_prefix_cache()
325
326
def sleep(self, level: int = 1):