diff --git a/python/sglang/srt/mem_cache/swa_radix_cache.py b/python/sglang/srt/mem_cache/swa_radix_cache.py
index 764def85c..928b207d8 100644
--- a/python/sglang/srt/mem_cache/swa_radix_cache.py
+++ b/python/sglang/srt/mem_cache/swa_radix_cache.py
@@ -449,11 +449,13 @@ class SWARadixCache(BasePrefixCache):
 
         if self.page_size != 1:
             page_aligned_len = actual_kv_len // self.page_size * self.page_size
-            page_aligned_kv_indices = kv_indices[:page_aligned_len].clone()
+            page_aligned_kv_indices = kv_indices[:page_aligned_len].to(
+                dtype=torch.int64, copy=True
+            )
             self.token_to_kv_pool_allocator.free(kv_indices[page_aligned_len:])
         else:
             page_aligned_len = actual_kv_len
-            page_aligned_kv_indices = kv_indices.clone()
+            page_aligned_kv_indices = kv_indices.to(dtype=torch.int64, copy=True)
             if self.is_eagle:
                 self.token_to_kv_pool_allocator.free(kv_indices[page_aligned_len:])
 
@@ -502,10 +504,12 @@ class SWARadixCache(BasePrefixCache):
 
         if self.page_size != 1:
             page_aligned_len = actual_kv_len // self.page_size * self.page_size
-            page_aligned_kv_indices = kv_indices[:page_aligned_len].clone()
+            page_aligned_kv_indices = kv_indices[:page_aligned_len].to(
+                dtype=torch.int64, copy=True
+            )
         else:
             page_aligned_len = actual_kv_len
-            page_aligned_kv_indices = kv_indices.clone()
+            page_aligned_kv_indices = kv_indices.to(dtype=torch.int64, copy=True)
 
         # For EAGLE, the page_aligned_len is for the bigram key, the normal key len should +1
         page_aligned_token_len = (