[Intel]Add 'intel_xpu' attention backend for llama4

python/sglang/srt/server_args.py

@@ -984,7 +984,8 @@ def _handle_model_specific_adjustments(self):
                 "aiter",
                 "triton",
                 "trtllm_mha",
-            }, "fa3, aiter, triton, or trtllm_mha is required for Llama4 model"
+                "intel_xpu",
+            }, "fa3, aiter, triton, trtllm_mha or intel_xpu is required for Llama4 model"
             if is_sm100_supported() and self.attention_backend is None:
                 self.attention_backend = "trtllm_mha"
                 logger.warning(

python/sglang/srt/utils/common.py

@@ -1831,7 +1831,8 @@ def get_device_capability(device_id: int = 0) -> Tuple[int, int]:
         major, minor, *_ = torch.xpu.get_device_capability(device_id)["version"].split(
             "."
         )
-        major, minor = int(major), int(minor)
+        # Currently XPU version does not contain capability information.
+        major, minor = None, None
 
     if hasattr(torch, "hpu") and torch.hpu.is_available():
         try: