未验证 提交 c74010e6 编写于 作者: D Double_V 提交者: GitHub

Merge pull request #4556 from tink2123/attn_score

fix attn score
......@@ -75,7 +75,7 @@ class AttentionHead(nn.Layer):
probs_step, axis=1)], axis=1)
next_input = probs_step.argmax(axis=1)
targets = next_input
probs = paddle.nn.functional.softmax(probs, axis=2)
return probs
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册