diff --git a/python/sglang/srt/mem_cache/hiradix_cache.py b/python/sglang/srt/mem_cache/hiradix_cache.py
index 718d186c3..9203e2db2 100644
--- a/python/sglang/srt/mem_cache/hiradix_cache.py
+++ b/python/sglang/srt/mem_cache/hiradix_cache.py
@@ -1287,9 +1287,17 @@ class HiRadixCache(RadixCache):
             self.evict_host(prefetch_length)
             host_indices = self.cache_controller.mem_pool_host.alloc(prefetch_length)
         if host_indices is None:
-            last_host_node.release_host()
-            # no sufficient host memory for prefetch
-            return
+            avaliable_size = self.cache_controller.mem_pool_host.available_size()
+            prefetch_length = avaliable_size - (avaliable_size % self.page_size)
+            if prefetch_length >= self.prefetch_threshold:
+                new_input_tokens = new_input_tokens[:prefetch_length]
+                host_indices = self.cache_controller.mem_pool_host.alloc(
+                    prefetch_length
+                )
+            else:
+                last_host_node.release_host()
+                # no sufficient host memory for prefetch
+                return
         operation = self.cache_controller.prefetch(
             req_id,
             host_indices,