xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 13:25:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[executor] feat: add supports_pp attr to executors (#21786)

Browse Source 
Signed-off-by: Haibin Lin <haibin.lin@bytedance.com>
This commit is contained in:
H 2025-08-03 03:04:45 -07:00 committed by GitHub
parent 7de45db9a5
commit 24d1dffbeb
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 17 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
vllm/engine/arg_utils.py
View File

@ -1490,14 +1490,18 @@ class EngineArgs:
and _warn_or_fallback("Engine in background thread")): and _warn_or_fallback("Engine in background thread")):
return False return False
if (self.pipeline_parallel_size > 1 if self.pipeline_parallel_size > 1:
and self.distributed_executor_backend supports_pp = getattr(self.distributed_executor_backend,
not in (ParallelConfig.distributed_executor_backend, "ray", 'supports_pp', False)
"mp", "external_launcher")): if not supports_pp and self.distributed_executor_backend not in (
name = "Pipeline Parallelism without Ray distributed executor " \ ParallelConfig.distributed_executor_backend, "ray", "mp",
"or multiprocessing executor or external launcher" "external_launcher"):
_raise_or_fallback(feature_name=name, recommend_to_remove=False) name = "Pipeline Parallelism without Ray distributed " \
return False "executor or multiprocessing executor or external " \
"launcher"
_raise_or_fallback(feature_name=name,
recommend_to_remove=False)
return False
# The platform may be supported on V1, but off by default for now. # The platform may be supported on V1, but off by default for now.
if not current_platform.default_v1( # noqa: SIM103 if not current_platform.default_v1( # noqa: SIM103

1
vllm/executor/executor_base.py
View File

@ -35,6 +35,7 @@ class ExecutorBase(ABC):
""" """
uses_ray: bool # whether the executor uses Ray for orchestration. uses_ray: bool # whether the executor uses Ray for orchestration.
supports_pp: bool = False # whether the executor supports PP
def __init__( def __init__(
self, self,

2
vllm/v1/executor/multiproc_executor.py
View File

@ -41,6 +41,8 @@ logger = init_logger(__name__)
class MultiprocExecutor(Executor): class MultiprocExecutor(Executor):
supports_pp: bool = True
def _init_executor(self) -> None: def _init_executor(self) -> None:
# Call self.shutdown at exit to clean up # Call self.shutdown at exit to clean up
# and ensure workers will be terminated. # and ensure workers will be terminated.

2
vllm/v1/executor/ray_distributed_executor.py
View File

@ -43,6 +43,8 @@ class FutureWrapper(Future):
class RayDistributedExecutor(RayDistributedExecutorV0, Executor): class RayDistributedExecutor(RayDistributedExecutorV0, Executor):
"""Ray distributed executor using Ray Compiled Graphs.""" """Ray distributed executor using Ray Compiled Graphs."""
supports_pp: bool = True
def _init_executor(self) -> None: def _init_executor(self) -> None:
super()._init_executor() super()._init_executor()