Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
DeepSpeech
提交
46088c0a
D
DeepSpeech
项目概览
PaddlePaddle
/
DeepSpeech
大约 2 年 前同步成功
通知
210
Star
8425
Fork
1598
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
245
列表
看板
标记
里程碑
合并请求
3
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
D
DeepSpeech
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
245
Issue
245
列表
看板
标记
里程碑
合并请求
3
合并请求
3
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
提交
46088c0a
编写于
9月 26, 2022
作者:
H
Hui Zhang
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
elimiate attn transpose
上级
f9e3eaa0
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
5 addition
and
3 deletion
+5
-3
paddlespeech/s2t/modules/attention.py
paddlespeech/s2t/modules/attention.py
+5
-3
未找到文件。
paddlespeech/s2t/modules/attention.py
浏览文件 @
46088c0a
...
@@ -271,7 +271,7 @@ class RelPositionMultiHeadedAttention(MultiHeadedAttention):
...
@@ -271,7 +271,7 @@ class RelPositionMultiHeadedAttention(MultiHeadedAttention):
and `head * d_k == size`
and `head * d_k == size`
"""
"""
q
,
k
,
v
=
self
.
forward_qkv
(
query
,
key
,
value
)
q
,
k
,
v
=
self
.
forward_qkv
(
query
,
key
,
value
)
q
=
q
.
transpose
([
0
,
2
,
1
,
3
])
# (batch, time1, head, d_k)
#
q = q.transpose([0, 2, 1, 3]) # (batch, time1, head, d_k)
# when export onnx model, for 1st chunk, we feed
# when export onnx model, for 1st chunk, we feed
# cache(1, head, 0, d_k * 2) (16/-1, -1/-1, 16/0 mode)
# cache(1, head, 0, d_k * 2) (16/-1, -1/-1, 16/0 mode)
...
@@ -302,9 +302,11 @@ class RelPositionMultiHeadedAttention(MultiHeadedAttention):
...
@@ -302,9 +302,11 @@ class RelPositionMultiHeadedAttention(MultiHeadedAttention):
p
=
p
.
transpose
([
0
,
2
,
1
,
3
])
# (batch, head, time1, d_k)
p
=
p
.
transpose
([
0
,
2
,
1
,
3
])
# (batch, head, time1, d_k)
# (batch, head, time1, d_k)
# (batch, head, time1, d_k)
q_with_bias_u
=
(
q
+
self
.
pos_bias_u
).
transpose
([
0
,
2
,
1
,
3
])
# q_with_bias_u = (q + self.pos_bias_u).transpose([0, 2, 1, 3])
q_with_bias_u
=
q
+
self
.
pos_bias_u
.
unsqueeze
(
1
)
# (batch, head, time1, d_k)
# (batch, head, time1, d_k)
q_with_bias_v
=
(
q
+
self
.
pos_bias_v
).
transpose
([
0
,
2
,
1
,
3
])
# q_with_bias_v = (q + self.pos_bias_v).transpose([0, 2, 1, 3])
q_with_bias_v
=
q
+
self
.
pos_bias_v
.
unsqueeze
(
1
)
# compute attention score
# compute attention score
# first compute matrix a and matrix c
# first compute matrix a and matrix c
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录