From 3c289e2104bf16b60607d626c6fc2e3eceb84f45 Mon Sep 17 00:00:00 2001
From: Zhen <295632982@qq.com>
Date: Fri, 23 May 2025 23:24:51 +0800
Subject: [PATCH] [performance_optim] reduce frequency of declaring
 attention_mask in Ascend NPU flash attention (#38278)

[performance_optim] reduce frequency of declaring attention_mask in ASCEND NPU flash attention
---
 .../integrations/npu_flash_attention.py            | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/src/transformers/integrations/npu_flash_attention.py b/src/transformers/integrations/npu_flash_attention.py
index bb515540d14..e32af9f4bc9 100644
--- a/src/transformers/integrations/npu_flash_attention.py
+++ b/src/transformers/integrations/npu_flash_attention.py
@@ -37,6 +37,8 @@ if SPARSE_MODE not in [TOP_LEFT_ALIGNED_CAUSAL_MASK_MODE, DOWN_RIGHT_ALIGNED_CAU
         "or 3 (down-right aligned causal mask)."
     )
 
+ATTN_MASK_NPU = None
+
 
 def is_npu_fa2_top_left_aligned_causal_mask():
     return SPARSE_MODE == TOP_LEFT_ALIGNED_CAUSAL_MASK_MODE if is_torch_npu_available() else False
@@ -171,7 +173,9 @@ def npu_flash_attn_func(
         head_num = q.shape[2]
         output = torch_npu.npu_fusion_attention(q, k, v, head_num, "BSND", keep_prob=keep_prob, scale=softmax_scale)[0]
     else:
-        attn_mask_npu = torch.triu(torch.ones([2048, 2048], device=q.device), diagonal=1).bool()
+        global ATTN_MASK_NPU
+        if ATTN_MASK_NPU is None:
+            ATTN_MASK_NPU = torch.triu(torch.ones([2048, 2048], device=q.device), diagonal=1).bool()
         head_num = q.shape[2]
         output = torch_npu.npu_fusion_attention(
             q,
@@ -181,7 +185,7 @@ def npu_flash_attn_func(
             "BSND",
             keep_prob=keep_prob,
             scale=softmax_scale,
-            atten_mask=attn_mask_npu,
+            atten_mask=ATTN_MASK_NPU,
             sparse_mode=SPARSE_MODE,
         )[0]
 
@@ -222,7 +226,9 @@ def npu_flash_attn_varlen_func(
             actual_seq_kvlen=tuple(cu_seqlens_k[1:].cpu().numpy().tolist()),
         )[0]
     else:
-        attn_mask_npu = torch.triu(torch.ones([2048, 2048], device=q.device), diagonal=1).bool()
+        global ATTN_MASK_NPU
+        if ATTN_MASK_NPU is None:
+            ATTN_MASK_NPU = torch.triu(torch.ones([2048, 2048], device=q.device), diagonal=1).bool()
         head_num = q.shape[1]
         output = torch_npu.npu_fusion_attention(
             q,
@@ -231,7 +237,7 @@ def npu_flash_attn_varlen_func(
             head_num,
             pse=None,
             padding_mask=None,
-            atten_mask=attn_mask_npu,
+            atten_mask=ATTN_MASK_NPU,
             scale=softmax_scale,
             keep_prob=keep_prob,
             input_layout="TND",