From 61d834a00ef31e925f93721de8703a3741bbf8da Mon Sep 17 00:00:00 2001
From: xiao <xiao1817836032@163.com>
Date: Wed, 24 Jun 2026 14:06:41 +0800
Subject: [PATCH] =?UTF-8?q?Add=20=E5=BC=80=E6=BA=90=E4=BB=93=E5=BA=93?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../FlashAttention关键算子迁移与优化.md             | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/基于AI Agent开发范式的国产GPU大模型推理算子库优化/FlashAttention关键算子迁移与优化.md b/基于AI Agent开发范式的国产GPU大模型推理算子库优化/FlashAttention关键算子迁移与优化.md
index 1aa5524..81ba709 100644
--- a/基于AI Agent开发范式的国产GPU大模型推理算子库优化/FlashAttention关键算子迁移与优化.md	
+++ b/基于AI Agent开发范式的国产GPU大模型推理算子库优化/FlashAttention关键算子迁移与优化.md	
@@ -255,6 +255,11 @@
 
 *   **Tensor Core:** 现代 GPU(含沐曦 C500)的矩阵乘法专用单元,要求维度对齐为 8 或 16 的倍数。
 
+### 5.5 开源仓库参考
+
+[GitHub - MetaX-MACA/flashattn · GitHub](https://github.com/MetaX-MACA/flashattn)
+
+> 链接内容可供用于学习 API、算子实现思路、benchmark 方法和优化策略。选手仍需根据 XPU-OJ 题包接口**自行实现**可提交的 `run_kernel(...)`
 
 ---