diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py index 91631cc3f..1660a2417 100644 --- a/python/sglang/srt/server_args.py +++ b/python/sglang/srt/server_args.py @@ -2465,8 +2465,8 @@ class ServerArgs: ) assert ( - is_cuda() or is_musa() - ), "Mamba extra_buffer is only supported on CUDA and MUSA devices with FLA backend" + is_cuda() or is_musa() or is_npu() + ), "Mamba extra_buffer is only supported on CUDA and MUSA and NPU devices with FLA backend" if self.speculative_num_draft_tokens is not None: assert ( self.mamba_track_interval >= self.speculative_num_draft_tokens