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 15, 06:00 UTC
[Mis] remove duplicate engine status checks (#19647)
Signed-off-by: googs1025 <googs1025@gmail.com>
CYJiang committed
Jun 17, 2025 at 23:17 UTC
5a1c2e15d847e30d9c72d60ba1e28dc4b89df23d
1 file changed
+3
-5
vllm/v1/engine/core_client.py
+3
-5
@@ -794,7 +794,6 @@ class AsyncMPClient(MPClient):
794
request_type: EngineCoreRequestType,
795
request: Any,
796
engine: Optional[CoreEngine] = None) -> Awaitable[Any]:
797
- self.ensure_alive()
797
if engine is None:
798
engine = self.core_engine
799
@@ -1059,7 +1058,7 @@ class DPAsyncMPClient(AsyncMPClient):
1058
self.reqs_in_flight.pop(req_id, None)
1059
1060
async def abort_requests_async(self, request_ids: list[str]) -> None:
1062
- if not request_ids:
1061
+ if not request_ids or self.resources.engine_dead:
1062
return
1063
1064
if len(request_ids) == 1:
@@ -1077,9 +1076,8 @@ class DPAsyncMPClient(AsyncMPClient):
1076
1077
async def _abort_requests(self, request_ids: list[str],
1078
engine: CoreEngine) -> None:
1080
- if not self.resources.engine_dead:
1081
- await self._send_input(EngineCoreRequestType.ABORT, request_ids,
1082
- engine)
1079
+ await self._send_input(EngineCoreRequestType.ABORT, request_ids,
1080
+ engine)
1081
1082
1083
class RayDPClient(DPAsyncMPClient):