diff --git a/vllm/v1/attention/backends/flash_attn.py b/vllm/v1/attention/backends/flash_attn.py
index a82f1bb20fba3..527b31153410b 100755
--- a/vllm/v1/attention/backends/flash_attn.py
+++ b/vllm/v1/attention/backends/flash_attn.py
@@ -266,7 +266,7 @@ class FlashAttentionMetadataBuilder(
         use_cascade = common_prefix_len > 0
 
         if use_cascade:
-            cu_prefix_query_lens = torch.tensor([0,num_actual_tokens],
+            cu_prefix_query_lens = torch.tensor([0, num_actual_tokens],
                                                 dtype=torch.int32,
                                                 device=self.runner.device)
             prefix_kv_lens = torch.tensor([common_prefix_len],