diff --git a/test/registered/cpu/test_activation.py b/test/registered/cpu/test_activation.py
index da870d47a..4ca87f45e 100644
--- a/test/registered/cpu/test_activation.py
+++ b/test/registered/cpu/test_activation.py
@@ -1,6 +1,7 @@
 import sys
 
 import pytest
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
diff --git a/test/registered/cpu/test_bmm.py b/test/registered/cpu/test_bmm.py
index bb014c085..dbcdb9e79 100644
--- a/test/registered/cpu/test_bmm.py
+++ b/test/registered/cpu/test_bmm.py
@@ -2,6 +2,7 @@ import itertools
 import unittest
 
 # TODO: use interface in cpu.py
+import sgl_kernel  # noqa: F401
 import torch
 import torch.nn as nn
 
diff --git a/test/registered/cpu/test_comm.py b/test/registered/cpu/test_comm.py
index 9999ceda6..181aaf239 100644
--- a/test/registered/cpu/test_comm.py
+++ b/test/registered/cpu/test_comm.py
@@ -5,6 +5,7 @@ import traceback
 import unittest
 from multiprocessing import Process
 
+import sgl_kernel  # noqa: F401
 import torch
 import torch.distributed as dist
 import torch.multiprocessing as mp
diff --git a/test/registered/cpu/test_decode.py b/test/registered/cpu/test_decode.py
index b6685c31c..dbbafbe16 100644
--- a/test/registered/cpu/test_decode.py
+++ b/test/registered/cpu/test_decode.py
@@ -1,5 +1,6 @@
 import unittest
 
+import sgl_kernel  # noqa: F401
 import torch
 from torch.nn.functional import scaled_dot_product_attention
 
diff --git a/test/registered/cpu/test_extend.py b/test/registered/cpu/test_extend.py
index 4400d45ed..e6bc055be 100644
--- a/test/registered/cpu/test_extend.py
+++ b/test/registered/cpu/test_extend.py
@@ -1,5 +1,6 @@
 import unittest
 
+import sgl_kernel  # noqa: F401
 import torch
 from torch.nn.functional import scaled_dot_product_attention
 
diff --git a/test/registered/cpu/test_gemm.py b/test/registered/cpu/test_gemm.py
index 45d8d043e..13e22065d 100644
--- a/test/registered/cpu/test_gemm.py
+++ b/test/registered/cpu/test_gemm.py
@@ -1,6 +1,7 @@
 import unittest
 
 # TODO: use interface in cpu.py
+import sgl_kernel  # noqa: F401
 import torch
 import torch.nn as nn
 
diff --git a/test/registered/cpu/test_mamba.py b/test/registered/cpu/test_mamba.py
index a3110b720..e2e933f01 100644
--- a/test/registered/cpu/test_mamba.py
+++ b/test/registered/cpu/test_mamba.py
@@ -1,6 +1,7 @@
 import sys
 
 import pytest
+import sgl_kernel  # noqa: F401
 import torch
 import torch.nn.functional as F
 from torch.nn.functional import softplus
diff --git a/test/registered/cpu/test_mla.py b/test/registered/cpu/test_mla.py
index 201705133..3fefb9be3 100644
--- a/test/registered/cpu/test_mla.py
+++ b/test/registered/cpu/test_mla.py
@@ -1,5 +1,6 @@
 import unittest
 
+import sgl_kernel  # noqa: F401
 import torch
 from torch.nn.functional import scaled_dot_product_attention
 
diff --git a/test/registered/cpu/test_moe.py b/test/registered/cpu/test_moe.py
index d6a3d1d26..4c903a5b7 100644
--- a/test/registered/cpu/test_moe.py
+++ b/test/registered/cpu/test_moe.py
@@ -3,6 +3,7 @@ import math
 import pytest
 
 # TODO: use interface in cpu.py
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.srt.layers.amx_utils import CPUQuantMethod
diff --git a/test/registered/cpu/test_norm.py b/test/registered/cpu/test_norm.py
index bcd30d3ee..7ee4eb8ad 100644
--- a/test/registered/cpu/test_norm.py
+++ b/test/registered/cpu/test_norm.py
@@ -2,6 +2,7 @@ import sys
 from typing import Optional, Tuple, Union
 
 import pytest
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.test.ci.ci_register import register_cpu_ci
diff --git a/test/registered/cpu/test_qkv_proj_with_rope.py b/test/registered/cpu/test_qkv_proj_with_rope.py
index c2aeb9800..f4f4ee490 100644
--- a/test/registered/cpu/test_qkv_proj_with_rope.py
+++ b/test/registered/cpu/test_qkv_proj_with_rope.py
@@ -1,5 +1,6 @@
 import unittest
 
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.srt.layers.quantization.fp8_utils import input_to_float8
diff --git a/test/registered/cpu/test_qwen3.py b/test/registered/cpu/test_qwen3.py
index a3d3ad43f..0467ba6e1 100644
--- a/test/registered/cpu/test_qwen3.py
+++ b/test/registered/cpu/test_qwen3.py
@@ -1,6 +1,7 @@
 import sys
 
 import pytest
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.srt.utils import is_host_cpu_arm64
diff --git a/test/registered/cpu/test_rope.py b/test/registered/cpu/test_rope.py
index 4a04011dc..b33590221 100644
--- a/test/registered/cpu/test_rope.py
+++ b/test/registered/cpu/test_rope.py
@@ -1,5 +1,6 @@
 import unittest
 
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.srt.layers.rotary_embedding import (
diff --git a/test/registered/cpu/test_shared_expert.py b/test/registered/cpu/test_shared_expert.py
index 749bf9914..5b18dbe69 100644
--- a/test/registered/cpu/test_shared_expert.py
+++ b/test/registered/cpu/test_shared_expert.py
@@ -2,6 +2,7 @@ import itertools
 import math
 import unittest
 
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.test.ci.ci_register import register_cpu_ci
diff --git a/test/registered/cpu/test_store_cache.py b/test/registered/cpu/test_store_cache.py
index 80bd94e06..bfca54310 100644
--- a/test/registered/cpu/test_store_cache.py
+++ b/test/registered/cpu/test_store_cache.py
@@ -1,6 +1,7 @@
 import sys
 
 import pytest
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.test.ci.ci_register import register_cpu_ci
diff --git a/test/registered/cpu/test_topk.py b/test/registered/cpu/test_topk.py
index 461940bf1..b855dc9d3 100644
--- a/test/registered/cpu/test_topk.py
+++ b/test/registered/cpu/test_topk.py
@@ -1,6 +1,7 @@
 import itertools
 import unittest
 
+import sgl_kernel  # noqa: F401
 import torch
 
 from sglang.srt.eplb.expert_location_dispatch import ExpertLocationDispatchInfo