[Platform] platform agnostic for EngineArgs initialization (#11225)

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

vllm/platforms/xpu.py

@@ -51,6 +51,10 @@ class XPUPlatform(Platform):
 
     @classmethod
     def check_and_update_config(cls, vllm_config: VllmConfig) -> None:
+        cache_config = vllm_config.cache_config
+        if cache_config and cache_config.block_size is None:
+            cache_config.block_size = 16
+
         # check and update model config
         model_config = vllm_config.model_config
         if model_config.dtype == torch.bfloat16: