diff --git a/vllm/v1/attention/backends/flash_attn.py b/vllm/v1/attention/backends/flash_attn.py
index d903bd89c..940dc7515 100755
--- a/vllm/v1/attention/backends/flash_attn.py
+++ b/vllm/v1/attention/backends/flash_attn.py
@@ -847,6 +847,7 @@ class FlashAttentionImpl(AttentionImpl):
             q_descale=q_descale,
             k_descale=k_descale,
             v_descale=v_descale,
+            num_splits=attn_metadata.max_num_splits,
         )
         # FA returns LSE in shape [ H, B ] but cp_lse_ag_out_rs wants [ B, H ]
         context_attn_out_cor, context_lse_cor = cp_lse_ag_out_rs(
@@ -876,6 +877,7 @@ class FlashAttentionImpl(AttentionImpl):
             q_descale=q_descale,
             k_descale=k_descale,
             v_descale=v_descale,
+            num_splits=attn_metadata.max_num_splits,
         )
         assert context_attn_out_cor.shape == query_attn_out.shape
         assert context_lse_cor.shape == query_lse.shape