From d05da62e9ccdf8e342b15bf6785d83224cc165af Mon Sep 17 00:00:00 2001 From: Yan Ma Date: Sun, 20 Sep 2026 14:00:31 +0800 Subject: [PATCH] [XPU] fix incorrect gdn kernel log for XPU path (#54871) Signed-off-by: Yan Ma --- vllm/model_executor/layers/mamba/gdn/qwen_gdn_linear_attn.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/vllm/model_executor/layers/mamba/gdn/qwen_gdn_linear_attn.py b/vllm/model_executor/layers/mamba/gdn/qwen_gdn_linear_attn.py index ccf1c60f0f..1ca07ad0cf 100644 --- a/vllm/model_executor/layers/mamba/gdn/qwen_gdn_linear_attn.py +++ b/vllm/model_executor/layers/mamba/gdn/qwen_gdn_linear_attn.py @@ -167,6 +167,8 @@ def _log_gdn_backend_decision( head_k_dim, ) return + elif current_platform.is_xpu(): + return chosen = { "flashinfer": "FlashInfer", @@ -524,6 +526,8 @@ class QwenGatedDeltaNetAttention(GatedDeltaNetAttention): self.gdn_decode_kernel = "triton" elif current_platform.is_cpu(): self.gdn_decode_kernel = "CPU" + elif current_platform.is_xpu(): + self.gdn_decode_kernel = "XPU" self.enable_fused_gdn_decode = self.gdn_decode_kernel == "cuda" logger.info_once("GDN decode kernel: %s", self.gdn_decode_kernel)