diff --git a/python/sglang/srt/managers/cache_controller.py b/python/sglang/srt/managers/cache_controller.py
index ac69c3e73..3d4905e50 100644
--- a/python/sglang/srt/managers/cache_controller.py
+++ b/python/sglang/srt/managers/cache_controller.py
@@ -22,13 +22,12 @@ from typing import TYPE_CHECKING, List, NamedTuple, Optional
 
 import torch
 
-from sglang.srt.mem_cache.hicache_storage import (
-    HiCacheStorageConfig,
-    HiCacheStorageExtraInfo,
-)
-
 if TYPE_CHECKING:
     from sglang.srt.mem_cache.allocator import BaseTokenToKVPoolAllocator
+    from sglang.srt.mem_cache.hicache_storage import (
+        HiCacheStorageConfig,
+        HiCacheStorageExtraInfo,
+    )
     from sglang.srt.mem_cache.memory_pool_host import HostKVCache
 
 from sglang.srt.distributed import (
@@ -926,9 +925,7 @@ class HiCacheController:
         for page in pages:
             self.host_mem_release_queue.put(page)
 
-    def _page_get_zero_copy(
-        self, operation, hash_values, host_indices, extra_info=None
-    ):
+    def _page_get_zero_copy(self, operation, hash_values, host_indices, extra_info=None):
         results = self.storage_backend.batch_get_v1(
             hash_values, host_indices, extra_info
         )
@@ -975,6 +972,8 @@ class HiCacheController:
             ]
             prev_completed_tokens = operation.completed_tokens
             # Get one batch token, and update the completed_tokens if succeed
+            from sglang.srt.mem_cache.hicache_storage import HiCacheStorageExtraInfo
+
             extra_info = HiCacheStorageExtraInfo(prefix_keys=prefix_keys)
             self.page_get_func(operation, batch_hashes, batch_host_indices, extra_info)
             # Check termination
@@ -1036,6 +1035,8 @@ class HiCacheController:
                     batch_tokens[i : i + self.page_size], last_hash
                 )
                 batch_hashes.append(last_hash)
+            from sglang.srt.mem_cache.hicache_storage import HiCacheStorageExtraInfo
+
             extra_info = HiCacheStorageExtraInfo(prefix_keys=prefix_keys)
             hit_page_num = self.storage_backend.batch_exists(batch_hashes, extra_info)
             hash_value.extend(batch_hashes[:hit_page_num])
@@ -1137,6 +1138,8 @@ class HiCacheController:
             ]
             # Set one batch token, and record if success.
             # todo: allow partial success
+            from sglang.srt.mem_cache.hicache_storage import HiCacheStorageExtraInfo
+
             extra_info = HiCacheStorageExtraInfo(prefix_keys=prefix_keys)
             success = self.page_set_func(batch_hashes, batch_host_indices, extra_info)
             if not success:
diff --git a/python/sglang/srt/mem_cache/hicache_storage.py b/python/sglang/srt/mem_cache/hicache_storage.py
index e76cfebd4..43644bd7c 100644
--- a/python/sglang/srt/mem_cache/hicache_storage.py
+++ b/python/sglang/srt/mem_cache/hicache_storage.py
@@ -1,14 +1,18 @@
+from __future__ import annotations
+
 import hashlib
 import logging
 import os
 from abc import ABC, abstractmethod
 from dataclasses import dataclass
-from typing import Any, List, Optional
+from typing import TYPE_CHECKING, Any, List, Optional
 
 import torch
 
 from sglang.srt.environ import envs
-from sglang.srt.mem_cache.memory_pool_host import HostKVCache
+
+if TYPE_CHECKING:
+    from sglang.srt.mem_cache.memory_pool_host import HostKVCache
 
 logger = logging.getLogger(__name__)
 
diff --git a/python/sglang/srt/mem_cache/hiradix_cache.py b/python/sglang/srt/mem_cache/hiradix_cache.py
index a48e49094..aa4d1241f 100644
--- a/python/sglang/srt/mem_cache/hiradix_cache.py
+++ b/python/sglang/srt/mem_cache/hiradix_cache.py
@@ -34,11 +34,6 @@ from sglang.srt.mem_cache.memory_pool import (
     MLATokenToKVPool,
     NSATokenToKVPool,
 )
-from sglang.srt.mem_cache.memory_pool_host import (
-    MHATokenToKVPoolHost,
-    MLATokenToKVPoolHost,
-    NSATokenToKVPoolHost,
-)
 from sglang.srt.mem_cache.radix_cache import (
     RadixCache,
     RadixKey,
@@ -136,6 +131,12 @@ class HiRadixCache(RadixCache):
                 )
         self.kv_cache = params.token_to_kv_pool_allocator.get_kvcache()
 
+        from sglang.srt.mem_cache.memory_pool_host import (
+            MHATokenToKVPoolHost,
+            MLATokenToKVPoolHost,
+            NSATokenToKVPoolHost,
+        )
+
         if isinstance(self.kv_cache, MHATokenToKVPool):
             self.token_to_kv_pool_host = MHATokenToKVPoolHost(
                 self.kv_cache,
diff --git a/test/registered/unit/mem_cache/test_cp_hicache_metadata.py b/test/registered/unit/mem_cache/test_cp_hicache_metadata.py
index 49fa4409e..0f61eaa1d 100644
--- a/test/registered/unit/mem_cache/test_cp_hicache_metadata.py
+++ b/test/registered/unit/mem_cache/test_cp_hicache_metadata.py
@@ -19,6 +19,24 @@ from sglang.test.test_utils import CustomTestCase
 register_cpu_ci(est_time=2, suite="stage-a-test-cpu")
 
 
+class TestCpHiCacheImports(CustomTestCase):
+    def test_cp_hicache_public_imports_without_sgl_kernel(self):
+        import subprocess
+
+        subprocess.run(
+            [
+                sys.executable,
+                "-c",
+                "from sglang.srt.mem_cache.hiradix_cache import HiRadixCache, CpHiCacheNodeMetadata; "
+                "from sglang.srt.managers.cache_controller import HiCacheController; "
+                "print('OK')",
+            ],
+            check=True,
+            capture_output=True,
+            text=True,
+        )
+
+
 class TestCpHiCacheNodeMetadata(CustomTestCase):
     def test_split_zero_len_moves_all_positions_to_child(self):
         metadata = CpHiCacheNodeMetadata(