Xpu/vllm 0.26 fixes (#5476)

Signed-off-by: Joshna Medisetty <joshna.medisetty@intel.com>
Co-authored-by: Chendi.Xue <chendi.xue@intel.com>

.buildkite/intel/pipeline-intel.yml

@@ -10,7 +10,7 @@ steps:
           DOCKER_BUILDKIT: "1"
           # Buildkite will automatically replace this with the actual commit hash
           VLLM_IMAGE_TAG: "${BUILDKITE_COMMIT}"
-          VLLM_VERSION: "v0.25.0"
+          VLLM_VERSION: "v0.26.0"
         priority: 100
         timeout_in_minutes: 60
         soft_fail: false

vllm_omni/platforms/interface.py

@@ -109,6 +109,7 @@ class OmniPlatform(Platform):
         cls,
         selected_backend: str | None,
         head_size: int,
+        allow_trtllm_default: bool = False,
     ) -> str:
         """Get the diffusion attention backend class path for this platform.
 
@@ -119,6 +120,7 @@ class OmniPlatform(Platform):
             selected_backend: User-selected backend name (e.g., "FLASH_ATTN",
                 "TORCH_SDPA", "SAGE_ATTN"). If None, uses platform default.
             head_size: Attention head size.
+            allow_trtllm_default: Whether TRTLLM may be chosen as the default.
 
         Returns:
             Fully qualified class path of the selected backend.

vllm_omni/platforms/xpu/platform.py

@@ -44,7 +44,9 @@ class XPUOmniPlatform(OmniPlatform, XPUPlatform):
         cls,
         selected_backend: str | None,
         head_size: int,
+        allow_trtllm_default: bool = False,
     ) -> str:
+        # XPU has no TRTLLM backend; arg accepted for signature parity but unused.
         compute_capability = torch.xpu.get_device_capability()
         # Intel Max 1100 and 1550 will not support flash_attn currently
         flash_attn_supported = compute_capability["architecture"] not in [13136561920]