diff --git a/python/sglang/srt/mem_cache/chunk_cache.py b/python/sglang/srt/mem_cache/chunk_cache.py
index 54626dffd..34cd2083f 100644
--- a/python/sglang/srt/mem_cache/chunk_cache.py
+++ b/python/sglang/srt/mem_cache/chunk_cache.py
@@ -27,6 +27,10 @@ class ChunkCache(BasePrefixCache):
         self.req_to_token_pool = req_to_token_pool
         self.token_to_kv_pool_allocator = token_to_kv_pool_allocator
         self.page_size = page_size
+        if self.token_to_kv_pool_allocator:
+            self.device = self.token_to_kv_pool_allocator.device
+        else:
+            self.device = torch.device("cpu")
 
     # NOTE (csy): this is to determine if a cache has prefix matching feature.
     # Chunk cache always return True to indicate no prefix matching.