diff --git a/python/sglang/srt/compilation/weak_ref_tensor.py b/python/sglang/srt/compilation/weak_ref_tensor.py
index 3564849c7..21d5f5d10 100644
--- a/python/sglang/srt/compilation/weak_ref_tensor.py
+++ b/python/sglang/srt/compilation/weak_ref_tensor.py
@@ -2,9 +2,9 @@ from typing import Any, Union
 
 import torch
 
-from sglang.srt.utils.common import is_cuda, is_hip, is_npu
+from sglang.srt.utils.common import is_cuda, is_hip, is_musa, is_npu
 
-if is_cuda() or is_hip():
+if is_cuda() or is_hip() or is_musa():
     from sgl_kernel import weak_ref_tensor
 elif is_npu():
     from torch_npu._C import _weak_ref_tensor as weak_ref_tensor
diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py
index 3b4659631..a638bf456 100644
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -51,7 +51,6 @@ from sglang.srt.utils.common import (
     is_hip,
     is_hopper_with_cuda_12_3,
     is_mps,
-    is_musa,
     is_no_spec_infer_or_topk_one,
     is_npu,
     is_remote_url,
@@ -1075,8 +1074,8 @@ class ServerArgs:
         # 5. Pipeline parallelism
         if self.pp_size > 1:
             self.disable_piecewise_cuda_graph = True
-        # 6. Non-CUDA hardware (AMD, NPU, CPU, MPS, MUSA, XPU, etc.)
-        if is_hip() or is_npu() or is_cpu() or is_mps() or is_musa() or is_xpu():
+        # 6. Non-CUDA hardware (AMD, NPU, CPU, MPS, XPU, etc.)
+        if is_hip() or is_npu() or is_cpu() or is_mps() or is_xpu():
             self.disable_piecewise_cuda_graph = True
         # 7. MoE A2A backend
         if self.moe_a2a_backend != "none":