biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Feat][EPLB] A novel static EPLB placement strategy for MoE models. (#23745)

Browse Source 
Signed-off-by: bruceszchen <bruceszchen@tencent.com>
Signed-off-by: Chen Bruce <bruceszchen@tencent.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Chen Bruce <cszwwdz@vip.qq.com>
Co-authored-by: lemon412 <lemon412@foxmail.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Chen Bruce
2025-09-16 18:55:16 +08:00
committed by GitHub
parent 27fcfe7bcf
commit 7ea5c73ad7
4 changed files with 265 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/engine/arg_utils.py
View File

@@ -34,6 +34,7 @@ from vllm.config import (BlockSize, CacheConfig, CacheDType, CompilationConfig,
SpeculativeConfig, TaskOption, TokenizerMode,
VllmConfig, get_attr_docs)
from vllm.config.multimodal import MMCacheType, MultiModalConfig
from vllm.config.parallel import ExpertPlacementStrategy
from vllm.config.utils import get_field
from vllm.logger import init_logger
from vllm.platforms import CpuArchEnum, current_platform
@@ -328,6 +329,8 @@ class EngineArgs:
enable_expert_parallel: bool = ParallelConfig.enable_expert_parallel
eplb_config: EPLBConfig = get_field(ParallelConfig, "eplb_config")
enable_eplb: bool = ParallelConfig.enable_eplb
expert_placement_strategy: ExpertPlacementStrategy = \
ParallelConfig.expert_placement_strategy
num_redundant_experts: int = EPLBConfig.num_redundant_experts
eplb_window_size: int = EPLBConfig.window_size
eplb_step_interval: int = EPLBConfig.step_interval
@@ -696,6 +699,9 @@ class EngineArgs:
**parallel_kwargs["enable_eplb"])
parallel_group.add_argument("--eplb-config",
**parallel_kwargs["eplb_config"])
parallel_group.add_argument(
"--expert-placement-strategy",
**parallel_kwargs["expert_placement_strategy"])
parallel_group.add_argument(
"--num-redundant-experts",
type=int,
@@ -1335,6 +1341,7 @@ class EngineArgs:
enable_expert_parallel=self.enable_expert_parallel,
enable_eplb=self.enable_eplb,
eplb_config=self.eplb_config,
expert_placement_strategy=self.expert_placement_strategy,
max_parallel_loading_workers=self.max_parallel_loading_workers,
disable_custom_all_reduce=self.disable_custom_all_reduce,
ray_workers_use_nsight=self.ray_workers_use_nsight,