From a4730c1b4fa297848aeb8b81a36c5158b3845e9c Mon Sep 17 00:00:00 2001
From: Chaojun Zhang <chaojun.zhang@intel.com>
Date: Wed, 12 Nov 2025 18:20:55 +0800
Subject: [PATCH] [XPU]Fix crash due to removed VLLM_USE_V1 attribute (#28520)

Signed-off-by: chaojun-zhang <chaojun.zhang@intel.com>
---
 vllm/platforms/xpu.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/vllm/platforms/xpu.py b/vllm/platforms/xpu.py
index 684d6d9a6b57..359eafc66445 100644
--- a/vllm/platforms/xpu.py
+++ b/vllm/platforms/xpu.py
@@ -65,7 +65,6 @@ class XPUPlatform(Platform):
 
         if use_sparse:
             raise NotImplementedError("Sparse Attention is not supported on XPU.")
-        use_v1 = envs.VLLM_USE_V1
         if not use_v1:
             raise ValueError("XPU backend only supports V1.")
         if selected_backend == AttentionBackendEnum.TRITON_ATTN:
@@ -115,7 +114,9 @@ class XPUPlatform(Platform):
     @classmethod
     def get_vit_attn_backend(
         cls, head_size: int, dtype: torch.dtype
-    ) -> AttentionBackendEnum:
+    ) -> "AttentionBackendEnum":
+        from vllm.attention.backends.registry import AttentionBackendEnum
+
         return AttentionBackendEnum.FLASH_ATTN
 
     @classmethod