[cpu/arm64] support run sglang on arm64 cpu (#14867)

python/sglang/srt/layers/activation.py

@@ -378,15 +378,3 @@ def get_cross_encoder_activation_function(config: PretrainedConfig):
     else:
         # adapt bge-reranker
         return nn.Identity()
-
-
-if not (
-    _is_cuda or _is_npu or (_is_cpu and _is_cpu_amx_available) or _is_hip or _is_xpu
-):
-    logger.info(
-        "sgl-kernel is not available on Non-NV, Non-AMD platforms or Non-AMX CPUs. Fallback to other kernel libraries."
-    )
-    from vllm.model_executor.layers.activation import (  # noqa: F401
-        GeluAndMul,
-        SiluAndMul,
-    )

python/sglang/srt/layers/layernorm.py

@@ -502,12 +502,3 @@ class Gemma3RMSNorm(MultiPlatformOp):
 
     def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.eps}"
-
-
-if not (
-    _is_cuda or _is_hip or _is_npu or (_is_cpu and _is_cpu_amx_available) or _is_xpu
-):
-    logger.info(
-        "sgl-kernel layernorm implementation is not available on current platform. Fallback to other kernel libraries."
-    )
-    from vllm.model_executor.layers.layernorm import GemmaRMSNorm, RMSNorm  # noqa: F401

python/sglang/srt/layers/rotary_embedding.py

@@ -116,7 +116,7 @@ class RotaryEmbedding(MultiPlatformOp):
 
         if (
             (not (_is_cuda or _is_npu) or self.head_size not in [64, 128, 256, 512])
-            and not (_is_cpu and _is_cpu_amx_available)
+            and not (_is_cpu)
             and not (_is_xpu)
         ):
             if _is_cuda or _is_hip: