[XPU]fallback to TRITON_ATTN for vit attn on xpu when use float32 dtype (#43759)

Signed-off-by: Yan Ma <yan.ma@intel.com>
Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>

vllm/platforms/xpu.py

@@ -110,6 +110,13 @@ class XPUPlatform(Platform):
         dtype: torch.dtype,
         backend: "AttentionBackendEnum | None" = None,
     ) -> "AttentionBackendEnum":
+        if dtype == torch.float32:
+            logger.warning_once(
+                "Flash Attention on XPU does not support float32 dtype. "
+                "Falling back to Triton Attention backend for vit attention."
+            )
+            return AttentionBackendEnum.TRITON_ATTN
+
         if backend is not None:
             assert backend in cls.get_supported_vit_attn_backends(), (
                 f"Backend {backend} is not supported for vit attention. "