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

[V1] DP scale-out (2/N): Decouple engine process management and comms (#15977)

Browse Source 
Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill
2025-05-13 10:48:21 -07:00
committed by GitHub
parent 0b217da646
commit 55aa7af994
10 changed files with 516 additions and 243 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
vllm/engine/arg_utils.py
View File

@@ -283,6 +283,9 @@ class EngineArgs:
pipeline_parallel_size: int = ParallelConfig.pipeline_parallel_size
tensor_parallel_size: int = ParallelConfig.tensor_parallel_size
data_parallel_size: int = ParallelConfig.data_parallel_size
data_parallel_size_local: Optional[int] = None
data_parallel_address: Optional[str] = None
data_parallel_rpc_port: Optional[int] = None
enable_expert_parallel: bool = ParallelConfig.enable_expert_parallel
max_parallel_loading_workers: Optional[
int] = ParallelConfig.max_parallel_loading_workers
@@ -596,6 +599,21 @@ class EngineArgs:
**parallel_kwargs["tensor_parallel_size"])
parallel_group.add_argument("--data-parallel-size", "-dp",
**parallel_kwargs["data_parallel_size"])
parallel_group.add_argument('--data-parallel-size-local',
'-dpl',
type=int,
help='Number of data parallel replicas '
'to run on this node.')
parallel_group.add_argument('--data-parallel-address',
'-dpa',
type=str,
help='Address of data parallel cluster '
'head-node.')
parallel_group.add_argument('--data-parallel-rpc-port',
'-dpp',
type=int,
help='Port for data parallel RPC '
'communication.')
parallel_group.add_argument(
"--enable-expert-parallel",
**parallel_kwargs["enable_expert_parallel"])
@@ -1019,10 +1037,30 @@ class EngineArgs:
# but we should not do this here.
placement_group = ray.util.get_current_placement_group()
# Local DP size defaults to global DP size if not set.
data_parallel_size_local = self.data_parallel_size if (
self.data_parallel_size_local
is None) else self.data_parallel_size_local
# DP address, used in multi-node case for torch distributed group
# and ZMQ sockets.
data_parallel_address = self.data_parallel_address if (
self.data_parallel_address
is not None) else ParallelConfig.data_parallel_master_ip
# This port is only used when there are remote data parallel engines,
# otherwise the local IPC transport is used.
data_parallel_rpc_port = self.data_parallel_rpc_port if (
self.data_parallel_rpc_port
is not None) else ParallelConfig.data_parallel_rpc_port
parallel_config = ParallelConfig(
pipeline_parallel_size=self.pipeline_parallel_size,
tensor_parallel_size=self.tensor_parallel_size,
data_parallel_size=self.data_parallel_size,
data_parallel_size_local=data_parallel_size_local,
data_parallel_master_ip=data_parallel_address,
data_parallel_rpc_port=data_parallel_rpc_port,
enable_expert_parallel=self.enable_expert_parallel,
max_parallel_loading_workers=self.max_parallel_loading_workers,
disable_custom_all_reduce=self.disable_custom_all_reduce,