From 05d7f7a166e2b4dd5a9e0445fd5993ea1a60d6ba Mon Sep 17 00:00:00 2001 From: Gabriel Wu <13583761+lucifer1004@users.noreply.github.com> Date: Tue, 20 Oct 2026 06:27:52 +0800 Subject: [PATCH] [Fix] attention: Honor selected FlashAttention implementation (#33007) --- python/sglang/srt/layers/attention/flashattention_backend.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/sglang/srt/layers/attention/flashattention_backend.py b/python/sglang/srt/layers/attention/flashattention_backend.py index be7088d7af..7177b153eb 100644 --- a/python/sglang/srt/layers/attention/flashattention_backend.py +++ b/python/sglang/srt/layers/attention/flashattention_backend.py @@ -1769,7 +1769,7 @@ class FlashAttentionBackend(AttentionBackend): else: metadata.fa_skip_cu_seqlens_q = cu_seqlens_q metadata.fa_skip_max_seqlen_q = max_seqlen_q - result = flash_attn_varlen_func( + result = self.flash_attn_varlen_func( q=q.contiguous().view(-1, layer.tp_q_head_num, layer.head_dim), k=k.view(-1, layer.tp_k_head_num, layer.head_dim), v=v.view(-1, layer.tp_v_head_num, layer.v_head_dim),