From 65494051201a0c8e521e668b85688f5c1de9bcd9 Mon Sep 17 00:00:00 2001
From: liu zhengxi <380185688@qq.com>
Date: Fri, 10 Dec 2021 20:35:05 +0800
Subject: [PATCH] Transfer MultiHeadAttention's matmul to v2 op (#36222)

* promote to v2

* alter
---
 python/paddle/nn/layer/transformer.py | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/python/paddle/nn/layer/transformer.py b/python/paddle/nn/layer/transformer.py
index 36bc836479..b0b6e62a60 100644
--- a/python/paddle/nn/layer/transformer.py
+++ b/python/paddle/nn/layer/transformer.py
@@ -402,9 +402,8 @@ class MultiHeadAttention(Layer):
             q, k, v, cache = self._prepare_qkv(query, key, value, cache)
 
         # scale dot product attention
-        # TODO(guosheng): use tensor.matmul, however it doesn't support `alpha`
-        product = layers.matmul(
-            x=q, y=k, transpose_y=True, alpha=self.head_dim**-0.5)
+        product = paddle.matmul(
+            x=q * (self.head_dim**-0.5), y=k, transpose_y=True)
         if attn_mask is not None:
             # Support bool or int mask
             attn_mask = _convert_attention_mask(attn_mask, product.dtype)
-- 
GitLab