From 599cce4d82383adc1cf250fb4e4fd792f19e68f0 Mon Sep 17 00:00:00 2001 From: Polisetty V R K Jyothendra Varma Date: Fri, 10 Apr 2026 12:40:00 +0530 Subject: [PATCH] [Intel GPU] import flash_attn functions from sgl_kernel only (#22438) --- python/sglang/srt/layers/attention/xpu_backend.py | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) diff --git a/python/sglang/srt/layers/attention/xpu_backend.py b/python/sglang/srt/layers/attention/xpu_backend.py index 77e773d88..4a40d25ee 100644 --- a/python/sglang/srt/layers/attention/xpu_backend.py +++ b/python/sglang/srt/layers/attention/xpu_backend.py @@ -20,11 +20,7 @@ if TYPE_CHECKING: from sglang.srt.model_executor.model_runner import ModelRunner from sgl_kernel import merge_state_v2 - -from sglang.jit_kernel.flash_attention import ( - flash_attn_varlen_func, - flash_attn_with_kvcache, -) +from sgl_kernel.flash_attn import flash_attn_varlen_func, flash_attn_with_kvcache class XPUAttentionBackend(AttentionBackend):