xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-15 11:55:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Platform] Allow platform use V1 Engine by default (#19792)

Browse Source 
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan 2025-06-18 21:03:36 +08:00 committed by GitHub
parent cca91a7a10
commit 257ab95439
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 20 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/engine/arg_utils.py
View File

@ -1303,7 +1303,7 @@ class EngineArgs:
# Skip this check if we are running on a non-GPU platform, # Skip this check if we are running on a non-GPU platform,
# or if the device capability is not available # or if the device capability is not available
# (e.g. in a Ray actor without GPUs). # (e.g. in a Ray actor without GPUs).
from vllm.platforms import CpuArchEnum, current_platform from vllm.platforms import current_platform
if (current_platform.is_cuda() if (current_platform.is_cuda()
and current_platform.get_device_capability() and current_platform.get_device_capability()
and current_platform.get_device_capability().major < 8): and current_platform.get_device_capability().major < 8):
@ -1445,14 +1445,10 @@ class EngineArgs:
_raise_or_fallback(feature_name=name, recommend_to_remove=False) _raise_or_fallback(feature_name=name, recommend_to_remove=False)
return False return False
# Non-[CUDA, TPU, x86 CPU] may be supported on V1, # The platform may be supported on V1, but off by default for now.
# but off by default for now. if not current_platform.default_v1( # noqa: SIM103
v0_hardware = not any( model_config=model_config) and _warn_or_fallback(
(current_platform.is_cuda_alike(), current_platform.is_tpu(), current_platform.device_name):
(current_platform.is_cpu()
and current_platform.get_cpu_architecture() == CpuArchEnum.X86)))
if v0_hardware and _warn_or_fallback( # noqa: SIM103
current_platform.device_name):
return False return False
############################################################# #############################################################

8
vllm/platforms/cpu.py
View File

@ -269,3 +269,11 @@ class CpuPlatform(Platform):
model configuration. model configuration.
""" """
return True return True
@classmethod
def default_v1(cls, model_config) -> bool:
"""Returns whether the current platform can use v1 by default for the
supplied model configuration.
"""
return cls.supports_v1(
model_config) and cls.get_cpu_architecture() == CpuArchEnum.X86

7
vllm/platforms/interface.py
View File

@ -479,6 +479,13 @@ class Platform:
""" """
return False return False
@classmethod
def default_v1(cls, model_config: ModelConfig) -> bool:
"""
Returns whether the current platform supports v1 by default.
"""
return cls.supports_v1(model_config)
@classmethod @classmethod
def use_custom_allreduce(cls) -> bool: def use_custom_allreduce(cls) -> bool:
""" """