diff --git a/python/cudnn/deepseek_sparse_attention/indexer_forward/_interface_sm90.py b/python/cudnn/deepseek_sparse_attention/indexer_forward/_interface_sm90.py
index b96474668..c83d4b5ae 100644
--- a/python/cudnn/deepseek_sparse_attention/indexer_forward/_interface_sm90.py
+++ b/python/cudnn/deepseek_sparse_attention/indexer_forward/_interface_sm90.py
@@ -47,8 +47,8 @@ def indexer_fwd(
     w: torch.Tensor,
     ratio: int = 4,
     qhead_per_kv_head: Optional[int] = None,
-    out: Optional[torch.Tensor] = None,
     sm_scale: float = 1.0,
+    out: Optional[torch.Tensor] = None,
     cu_seqlens_q: Optional[torch.Tensor] = None,
     cu_seqlens_k: Optional[torch.Tensor] = None,
     max_seqlen_q: Optional[int] = None,
diff --git a/python/cudnn/deepseek_sparse_attention/indexer_forward/api.py b/python/cudnn/deepseek_sparse_attention/indexer_forward/api.py
index ccf4b722d..bf3a0b1f9 100644
--- a/python/cudnn/deepseek_sparse_attention/indexer_forward/api.py
+++ b/python/cudnn/deepseek_sparse_attention/indexer_forward/api.py
@@ -229,6 +229,7 @@ def indexer_forward_wrapper(
     q_stage: int = 2,
     kv_stage: int = 4,
     sm_scale: float = 1.0,
+    out: Optional[torch.Tensor] = None,
     stream: Optional[cuda.CUstream] = None,
     cu_seqlens_q: Optional[torch.Tensor] = None,
     cu_seqlens_k: Optional[torch.Tensor] = None,
@@ -267,6 +268,7 @@ def indexer_forward_wrapper(
             ratio=ratio,
             qhead_per_kv_head=qhead_per_kv_head,
             sm_scale=sm_scale,
+            out=out,
             cu_seqlens_q=cu_seqlens_q,
             cu_seqlens_k=cu_seqlens_k,
             max_seqlen_q=max_seqlen_q,
@@ -286,6 +288,7 @@ def indexer_forward_wrapper(
         w,
         ratio=ratio,
         qhead_per_kv_head=qhead_per_kv_head,
+        out=out,
         m_block_size=m_block_size,
         n_block_size=n_block_size,
         num_threads=384,