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

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

Author: wangxiyuan

vllm/engine/arg_utils.py

@@ -1303,7 +1303,7 @@ def _is_v1_supported_oracle(self, model_config: ModelConfig) -> bool:
         # Skip this check if we are running on a non-GPU platform,
         # or if the device capability is not available
         # (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()
                 and current_platform.get_device_capability()
                 and current_platform.get_device_capability().major < 8):
@@ -1445,14 +1445,10 @@ def _is_v1_supported_oracle(self, model_config: ModelConfig) -> bool:
             _raise_or_fallback(feature_name=name, recommend_to_remove=False)
             return False
 
-        # Non-[CUDA, TPU, x86 CPU] may be supported on V1,
-        # but off by default for now.
-        v0_hardware = not any(
-            (current_platform.is_cuda_alike(), current_platform.is_tpu(),
-             (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):
+        # The platform may be supported on V1, but off by default for now.
+        if not current_platform.default_v1(  # noqa: SIM103
+                model_config=model_config) and _warn_or_fallback(
+                    current_platform.device_name):
             return False
         #############################################################
 

vllm/platforms/cpu.py

@@ -269,3 +269,11 @@ def supports_v1(cls, model_config) -> bool:
         model configuration.
         """
         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

vllm/platforms/interface.py

@@ -479,6 +479,13 @@ def supports_v1(cls, model_config: ModelConfig) -> bool:
         """
         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
     def use_custom_allreduce(cls) -> bool:
         """