From 0fc9598b3e2adb0ddf3c71ae788cb582263235ee Mon Sep 17 00:00:00 2001 From: silencejade <222910030+silencejade@users.noreply.github.com> Date: Mon, 11 May 2026 09:40:44 +0800 Subject: [PATCH] [NPU] Support radix-cache with mamba-extra-buffer for Qwen3.5 (#23891) --- python/sglang/srt/server_args.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) 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