From a5a64a311a39b153d1e4d3d6bcb67e77cdc9aeae Mon Sep 17 00:00:00 2001
From: Hanming Lu <69857889+hanming-lu@users.noreply.github.com>
Date: Sat, 23 May 2026 03:25:57 -0700
Subject: [PATCH] [Spec] trtllm mha supports overlap plan stream (#25925)

---
 python/sglang/srt/layers/attention/trtllm_mha_backend.py | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/python/sglang/srt/layers/attention/trtllm_mha_backend.py b/python/sglang/srt/layers/attention/trtllm_mha_backend.py
index e68bcb95e..b74a11846 100644
--- a/python/sglang/srt/layers/attention/trtllm_mha_backend.py
+++ b/python/sglang/srt/layers/attention/trtllm_mha_backend.py
@@ -537,6 +537,11 @@ class TRTLLMHAAttnBackend(FlashInferAttnBackend):
             self._copy_swa_page_table(metadata, page_indices, max_seq_pages)
         self.forward_metadata = metadata
 
+    def update_verify_buffers_to_fill_after_draft(
+        self, spec_info: SpecInput, cuda_graph_bs: Optional[int]
+    ):
+        pass
+
     def get_cuda_graph_seq_len_fill_value(self) -> int:
         """Get the fill value for sequence lengths in CUDA graph."""
         return 1