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 14, 05:30 UTC
scheduler.py: Update the name of the default scheduler. (#26758)
Signed-off-by: Ryan Li <ryanli@ryanli.org>
Ryan Li committed
Oct 14, 2025 at 14:52 UTC
481545b397572ff460f4eee747df2066cde9501c
2 files changed
+6
-11
vllm/config/scheduler.py
+6
-6
@@ -131,12 +131,12 @@ class SchedulerConfig:
131
some image tokens can be scheduled (like TTTTIIIII, leaving IIIII),
132
it will be scheduled as TTTT in one step and IIIIIIIIII in the next."""
133
134
- # scheduler class or path. "vllm.core.scheduler.Scheduler" (default)
135
- # or "mod.custom_class".
136
- scheduler_cls: str | type[object] = "vllm.core.scheduler.Scheduler"
137
- """The scheduler class to use. "vllm.core.scheduler.Scheduler" is the
138
- default scheduler. Can be a class directly or the path to a class of form
139
- "mod.custom_class"."""
134
+ # scheduler class or path. "vllm.v1.core.sched.scheduler.Scheduler"
135
+ # (default) or "mod.custom_class".
136
+ scheduler_cls: str | type[object] = "vllm.v1.core.sched.scheduler.Scheduler"
137
+ """The scheduler class to use. "vllm.v1.core.sched.scheduler.Scheduler" is
138
+ the default scheduler. Can be a class directly or the path to a class of
139
+ form "mod.custom_class"."""
140
141
disable_hybrid_kv_cache_manager: bool = False
142
"""If set to True, KV cache manager will allocate the same size of KV cache
vllm/engine/arg_utils.py
-5
@@ -1742,11 +1742,6 @@ class EngineArgs:
1742
self.enable_prefix_caching = incremental_prefill_supported
1743
logger.info("(%s) prefix caching by default", action)
1744
1745
- # V1 should use the new scheduler by default.
1746
- # Swap it only if this arg is set to the original V0 default
1747
- if self.scheduler_cls == EngineArgs.scheduler_cls:
1748
- self.scheduler_cls = "vllm.v1.core.sched.scheduler.Scheduler"
1749
-
1745
# When no user override, set the default values based on the usage
1746
# context.
1747
# Use different default values for different hardware.