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, 02:00 UTC
[Bugfix] Fix wrong CLI defaults for dynamic `SchedulerConfig` fields (#28872)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Nov 18, 2025 at 12:30 UTC
bf9e1e8767fb4d1143b7e042ed940b84ef031c66
2 files changed
+17
-5
tests/entrypoints/openai/test_enable_force_include_usage.py
+2
-2
@@ -17,7 +17,7 @@ def chat_server_with_force_include_usage(request): # noqa: F811
17
"128",
18
"--enforce-eager",
19
"--max-num-seqs",
20
- "1",
20
+ "4",
21
"--enable-force-include-usage",
22
"--port",
23
"55857",
@@ -78,7 +78,7 @@ def transcription_server_with_force_include_usage():
78
"--dtype",
79
"bfloat16",
80
"--max-num-seqs",
81
- "1",
81
+ "4",
82
"--enforce-eager",
83
"--enable-force-include-usage",
84
"--gpu-memory-utilization",
vllm/engine/arg_utils.py
+15
-3
@@ -1046,10 +1046,18 @@ class EngineArgs:
1046
description=SchedulerConfig.__doc__,
1047
)
1048
scheduler_group.add_argument(
1049
- "--max-num-batched-tokens", **scheduler_kwargs["max_num_batched_tokens"]
1049
+ "--max-num-batched-tokens",
1050
+ **{
1051
+ **scheduler_kwargs["max_num_batched_tokens"],
1052
+ "default": None,
1053
+ },
1054
)
1055
scheduler_group.add_argument(
1052
- "--max-num-seqs", **scheduler_kwargs["max_num_seqs"]
1056
+ "--max-num-seqs",
1057
+ **{
1058
+ **scheduler_kwargs["max_num_seqs"],
1059
+ "default": None,
1060
+ },
1061
)
1062
scheduler_group.add_argument(
1063
"--max-num-partial-prefills", **scheduler_kwargs["max_num_partial_prefills"]
@@ -1071,7 +1079,11 @@ class EngineArgs:
1079
"--scheduling-policy", **scheduler_kwargs["policy"]
1080
)
1081
scheduler_group.add_argument(
1074
- "--enable-chunked-prefill", **scheduler_kwargs["enable_chunked_prefill"]
1082
+ "--enable-chunked-prefill",
1083
+ **{
1084
+ **scheduler_kwargs["enable_chunked_prefill"],
1085
+ "default": None,
1086
+ },
1087
)
1088
scheduler_group.add_argument(
1089
"--disable-chunked-mm-input", **scheduler_kwargs["disable_chunked_mm_input"]