support qwen2 running on ascend npu device (#7022)

Co-authored-by: 刁莹煜 <diaoyingyu1@hisilicon.com>

python/sglang/srt/layers/layernorm.py

@@ -20,10 +20,11 @@ import torch
 import torch.nn as nn
 
 from sglang.srt.custom_op import CustomOp
-from sglang.srt.utils import get_bool_env_var, is_cuda, is_hip
+from sglang.srt.utils import get_bool_env_var, is_cuda, is_hip, is_npu
 
 _is_cuda = is_cuda()
 _is_hip = is_hip()
+_is_npu = is_npu()
 _use_aiter = get_bool_env_var("SGLANG_USE_AITER") and _is_hip
 
 if _is_cuda:
@@ -187,7 +188,7 @@ class Gemma3RMSNorm(nn.Module):
         return f"{tuple(self.weight.shape)}, eps={self.eps}"
 
 
-if not (_is_cuda or _is_hip):
+if not (_is_cuda or _is_hip or _is_npu):
     logger.info(
         "sgl-kernel layernorm implementation is not available on current platform. Fallback to other kernel libraries."
     )