Add cache to cuda get_device_capability (#19436)

Signed-off-by: mgoin <mgoin64@gmail.com>

Author: mgoin

vllm/platforms/cuda.py

@@ -6,7 +6,7 @@
 
 import os
 from datetime import timedelta
-from functools import wraps
+from functools import cache, wraps
 from typing import TYPE_CHECKING, Callable, Optional, TypeVar, Union
 
 import torch
@@ -389,6 +389,7 @@ def stateless_init_device_torch_dist_pg(
 class NvmlCudaPlatform(CudaPlatformBase):
 
     @classmethod
+    @cache
     @with_nvml_context
     def get_device_capability(cls,
                               device_id: int = 0
@@ -486,6 +487,7 @@ def log_warnings(cls):
 class NonNvmlCudaPlatform(CudaPlatformBase):
 
     @classmethod
+    @cache
     def get_device_capability(cls, device_id: int = 0) -> DeviceCapability:
         major, minor = torch.cuda.get_device_capability(device_id)
         return DeviceCapability(major=major, minor=minor)