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 12, 14:45 UTC
Scheduled removal of `ParallelConfig`'s direct child EPLB fields (#29324)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Nov 25, 2025 at 01:58 UTC
a4ad43ad5a819aabc7d9b48b46a7f11e2552befc
2 files changed
-74
vllm/config/parallel.py
-50
@@ -141,22 +141,6 @@ class ParallelConfig:
141
- "deepep_high_throughput": Use deepep high-throughput kernels
142
- "deepep_low_latency": Use deepep low-latency kernels
143
- "flashinfer_all2allv": Use flashinfer alltoallv kernels for mnnvl"""
144
- num_redundant_experts: int | None = None
145
- """`num_redundant_experts` is deprecated and has been replaced with
146
- `eplb_config.num_redundant_experts`. This will be removed in v0.12.0.
147
- Please use `eplb_config.num_redundant_experts` instead."""
148
- eplb_window_size: int | None = None
149
- """`eplb_window_size` is deprecated and has been replaced with
150
- `eplb_config.window_size`. This will be removed in v0.12.0.
151
- Please use `eplb_config.window_size` instead."""
152
- eplb_step_interval: int | None = None
153
- """`eplb_step_interval` is deprecated and has been replaced with
154
- `eplb_config.step_interval`. This will be removed in v0.12.0.
155
- Please use `eplb_config.step_interval` instead."""
156
- eplb_log_balancedness: bool | None = None
157
- """`eplb_log_balancedness` is deprecated and has been replaced with
158
- `eplb_config.log_balancedness`. This will be removed in v0.12.0.
159
- Please use `eplb_config.log_balancedness` instead."""
144
145
max_parallel_loading_workers: int | None = None
146
"""Maximum number of parallel loading workers when loading model
@@ -516,40 +500,6 @@ class ParallelConfig:
500
"--all2all-backend command-line argument instead."
501
)
502
519
- # Forward deprecated fields to their new location
520
- if self.num_redundant_experts is not None:
521
- self.eplb_config.num_redundant_experts = self.num_redundant_experts
522
- logger.warning_once(
523
- "num_redundant_experts is deprecated and has been replaced "
524
- "with eplb_config.num_redundant_experts. This will be removed "
525
- "in v0.12.0. Changing this field after initialization will "
526
- "have no effect."
527
- )
528
- if self.eplb_window_size is not None:
529
- self.eplb_config.window_size = self.eplb_window_size
530
- logger.warning_once(
531
- "eplb_window_size is deprecated and has been replaced "
532
- "with eplb_config.window_size. This will be removed "
533
- "in v0.12.0. Changing this field after initialization will "
534
- "have no effect."
535
- )
536
- if self.eplb_step_interval is not None:
537
- self.eplb_config.step_interval = self.eplb_step_interval
538
- logger.warning_once(
539
- "eplb_step_interval is deprecated and has been replaced "
540
- "with eplb_config.step_interval. This will be removed "
541
- "in v0.12.0. Changing this field after initialization will "
542
- "have no effect."
543
- )
544
- if self.eplb_log_balancedness is not None:
545
- self.eplb_config.log_balancedness = self.eplb_log_balancedness
546
- logger.warning_once(
547
- "eplb_log_balancedness is deprecated and has been replaced "
548
- "with eplb_config.log_balancedness. This will be removed "
549
- "in v0.12.0. Changing this field after initialization will "
550
- "have no effect."
551
- )
552
-
503
# Continue with the rest of the initialization
504
self.world_size = (
505
self.pipeline_parallel_size
vllm/engine/arg_utils.py
-24
@@ -855,30 +855,6 @@ class EngineArgs:
855
"--expert-placement-strategy",
856
**parallel_kwargs["expert_placement_strategy"],
857
)
858
- parallel_group.add_argument(
859
- "--num-redundant-experts",
860
- type=int,
861
- help="[DEPRECATED] --num-redundant-experts will be removed in v0.12.0.",
862
- deprecated=True,
863
- )
864
- parallel_group.add_argument(
865
- "--eplb-window-size",
866
- type=int,
867
- help="[DEPRECATED] --eplb-window-size will be removed in v0.12.0.",
868
- deprecated=True,
869
- )
870
- parallel_group.add_argument(
871
- "--eplb-step-interval",
872
- type=int,
873
- help="[DEPRECATED] --eplb-step-interval will be removed in v0.12.0.",
874
- deprecated=True,
875
- )
876
- parallel_group.add_argument(
877
- "--eplb-log-balancedness",
878
- action=argparse.BooleanOptionalAction,
879
- help="[DEPRECATED] --eplb-log-balancedness will be removed in v0.12.0.",
880
- deprecated=True,
881
- )
858
859
parallel_group.add_argument(
860
"--max-parallel-loading-workers",