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, 19:45 UTC
[bugfix] fixup warning message for plugged schedulers for v1 (#14700)
Signed-off-by: Joe Runde <Joseph.Runde@ibm.com>
Joe Runde committed
Mar 12, 2025 at 21:12 UTC
165290d357bf0b95a78a96a45cb7e0dd98d494de
1 file changed
+11
-4
vllm/v1/engine/core.py
+11
-4
@@ -22,6 +22,7 @@ from vllm.transformers_utils.config import (
22
from vllm.utils import (get_exception_traceback, resolve_obj_by_qualname,
23
zmq_socket_ctx)
24
from vllm.v1.core.kv_cache_utils import get_kv_cache_configs
25
+from vllm.v1.core.scheduler import Scheduler as V1Scheduler
26
from vllm.v1.core.scheduler import SchedulerOutput
27
from vllm.v1.engine import (EngineCoreOutputs, EngineCoreRequest,
28
EngineCoreRequestType, UtilityOutput)
@@ -67,15 +68,21 @@ class EngineCore:
68
69
# Setup scheduler.
70
if isinstance(vllm_config.scheduler_config.scheduler_cls, str):
71
+ Scheduler = resolve_obj_by_qualname(
72
+ vllm_config.scheduler_config.scheduler_cls)
73
+ else:
74
+ Scheduler = vllm_config.scheduler_config.scheduler_cls
75
+
76
+ # This warning can be removed once the V1 Scheduler interface is
77
+ # finalized and we can maintain support for scheduler classes that
78
+ # implement it
79
+ if Scheduler is not V1Scheduler:
80
logger.warning(
81
"Using configured V1 scheduler class %s. "
82
"This scheduler interface is not public and "
83
"compatibility may not be maintained.",
84
vllm_config.scheduler_config.scheduler_cls)
75
- Scheduler = resolve_obj_by_qualname(
76
- vllm_config.scheduler_config.scheduler_cls)
77
- else:
78
- Scheduler = vllm_config.scheduler_config.scheduler_cls
85
+
86
self.scheduler = Scheduler(
87
scheduler_config=vllm_config.scheduler_config,
88
model_config=vllm_config.model_config,