[bugfix] add supports_v1 platform interface (#15417)

Signed-off-by: Joe Runde <Joseph.Runde@ibm.com>
This commit is contained in:
Joe Runde
2025-03-25 15:00:32 -04:00
committed by GitHub
parent 5d8e1c9279
commit 5f063a80bd
5 changed files with 31 additions and 7 deletions
+6 -1
View File
@@ -20,8 +20,9 @@ from vllm.utils import import_pynvml
from .interface import DeviceCapability, Platform, PlatformEnum, _Backend
if TYPE_CHECKING:
from vllm.config import VllmConfig
from vllm.config import ModelConfig, VllmConfig
else:
ModelConfig = None
VllmConfig = None
logger = init_logger(__name__)
@@ -303,6 +304,10 @@ class CudaPlatformBase(Platform):
def supports_fp8(cls) -> bool:
return cls.has_device_capability(89)
@classmethod
def supports_v1(cls, model_config: ModelConfig) -> bool:
return True
# NVML utils
# Note that NVML is not affected by `CUDA_VISIBLE_DEVICES`,