From 2fc28056fafd3d5b9fdf57b0f166c5f795c01504 Mon Sep 17 00:00:00 2001
From: "songrui.771" <songrui.771@bytedance.com>
Date: Tue, 27 May 2025 16:14:07 +0800
Subject: [PATCH] fix performance regression due to Flash Attention duplicated
 code

---
 wan/distributed/xdit_context_parallel.py | 7 -------
 1 file changed, 7 deletions(-)

diff --git a/wan/distributed/xdit_context_parallel.py b/wan/distributed/xdit_context_parallel.py
index 2df66c4..b264d84 100644
--- a/wan/distributed/xdit_context_parallel.py
+++ b/wan/distributed/xdit_context_parallel.py
@@ -304,13 +304,6 @@ def usp_attn_forward(self,
             value=half(v),
             window_size=self.window_size)
 
-    x = xFuserLongContextAttention()(
-        None,
-        query=half(q),
-        key=half(k),
-        value=half(v),
-        window_size=self.window_size)
-
     # TODO: padding after attention.
     # x = torch.cat([x, x.new_zeros(b, s - x.size(1), n, d)], dim=1)