From 94ce940ff80f74787ab297d20a6bc828bfe2b454 Mon Sep 17 00:00:00 2001 From: Liangsheng Yin Date: Thu, 10 Sep 2026 21:03:03 -0700 Subject: [PATCH] [Bug] Include DSA variant in exact-bucket graph admission (#38992) --- .../srt/model_executor/runner/decode_cuda_graph_runner.py | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/python/sglang/srt/model_executor/runner/decode_cuda_graph_runner.py b/python/sglang/srt/model_executor/runner/decode_cuda_graph_runner.py index e7abb66e0..51463f6e4 100644 --- a/python/sglang/srt/model_executor/runner/decode_cuda_graph_runner.py +++ b/python/sglang/srt/model_executor/runner/decode_cuda_graph_runner.py @@ -705,6 +705,11 @@ class DecodeCudaGraphRunner(BaseCudaGraphRunner): cuda_graph_bs, stream_idx=get_current_stream_idx() if self.enable_pdmux else None, variant_label=self._resolve_lora_variant(forward_batch), + dsa_variant=( + self._resolve_dsa_variant(forward_batch) + if self.disable_padding + else None + ), ) is_bs_supported = (