diff --git a/python/sglang/srt/hardware_backend/npu/attention/ascend_hybrid_linear_attn_backend.py b/python/sglang/srt/hardware_backend/npu/attention/ascend_hybrid_linear_attn_backend.py index eca7ce9b9..3d1e7af81 100644 --- a/python/sglang/srt/hardware_backend/npu/attention/ascend_hybrid_linear_attn_backend.py +++ b/python/sglang/srt/hardware_backend/npu/attention/ascend_hybrid_linear_attn_backend.py @@ -131,6 +131,7 @@ class AscendMambaAttnBackendBase(MambaAttnBackendBase): spec_info: Optional[SpecInput], seq_lens_cpu: Optional[torch.Tensor], num_padding: Optional[int] = None, + in_capture: bool = False, ): # out_graph passes seq_lens_cpu=None at capture; mirror the base guard. if seq_lens_cpu is None: