From 72847b8800d7ede6deb9d6f8cb429c8097370fca Mon Sep 17 00:00:00 2001
From: will-jl944 <lujiafeng@baidu.com>
Date: Thu, 9 Jan 2025 18:46:08 +0800
Subject: [PATCH] [NPU] Adapt to new flash_attention_npu api

---
 paddlenlp/transformers/llama/fusion_ops.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/paddlenlp/transformers/llama/fusion_ops.py b/paddlenlp/transformers/llama/fusion_ops.py
index 4cb9101cc730..c2a95541fbbd 100644
--- a/paddlenlp/transformers/llama/fusion_ops.py
+++ b/paddlenlp/transformers/llama/fusion_ops.py
@@ -206,8 +206,8 @@ def fusion_flash_attention(
                 value_states,
                 None,
                 attention_mask,
-                [],
-                [],
+                None,
+                None,
                 0.0,
                 attention_mask is None,
                 True,