Add intel_amx backend for Radix Attention for CPU (#6408)

Co-authored-by: Chunyuan WU <chunyuan.wu@intel.com>
Co-authored-by: Thien Tran <gau.nernst@yahoo.com.sg>

python/sglang/srt/server_args.py

@@ -323,6 +323,11 @@ class ServerArgs:
             self.sampling_backend = "pytorch"
 
         # Set kernel backends
+        if self.device == "cpu":
+            if self.attention_backend is None:
+                self.attention_backend = "intel_amx"
+            self.sampling_backend = "pytorch"
+
         if self.sampling_backend is None:
             self.sampling_backend = (
                 "flashinfer" if is_flashinfer_available() else "pytorch"
@@ -993,6 +998,7 @@ class ServerArgs:
                 "fa3",
                 "flashmla",
                 "cutlass_mla",
+                "intel_amx",
             ],
             default=ServerArgs.attention_backend,
             help="Choose the kernels for attention layers.",