diff --git a/python/sglang/srt/layers/attention/flashattention_backend.py b/python/sglang/srt/layers/attention/flashattention_backend.py
index 365a0a54f..e1f7ea76f 100644
--- a/python/sglang/srt/layers/attention/flashattention_backend.py
+++ b/python/sglang/srt/layers/attention/flashattention_backend.py
@@ -322,10 +322,13 @@ class FlashAttentionBackend(AttentionBackend):
             torch.cumsum(seq_lens_in_batch, dim=0, dtype=torch.int32), (1, 0)
         )
 
-        metadata.page_table = self.req_to_token[
-            :, self.decode_cuda_graph_metadata["strided_indices"]
+        max_seq_pages = (metadata.max_seq_len_k + self.page_size - 1) // self.page_size
+        page_indices = self.req_to_token[
+            :, self.decode_cuda_graph_metadata["strided_indices"][:max_seq_pages]
         ]
-        metadata.page_table = metadata.page_table[req_pool_indices[:bs]]
+        page_indices = page_indices[req_pool_indices[:bs]] // self.page_size
+        metadata.page_table[:, :max_seq_pages].copy_(page_indices)
+        metadata.page_table[:, max_seq_pages:].fill_(0)
         self.forward_metadata = metadata
 
     def get_cuda_graph_seq_len_fill_value(self):