关于Transformer拼音输入法 #8
Description
UP你的视频最后我已经做出来了对应的demo在2个月前，可以联系kivvi3412@gmail.com讨论一下
我是用的是transformer 的encoder decoder模型，使用单张A100训练了一天，目前还有一些小问题可以探讨
这是我demo的输出
[上下文以及输入拼音] Context='', Pinyin='zhi'
--- Beam Search Top 5 Candidates ---
Rank 1: Score=-1.8559 | Result='只'
Rank 2: Score=-2.1351 | Result='直'
Rank 3: Score=-2.4737 | Result='制'
Rank 4: Score=-2.4852 | Result='知'
Rank 5: Score=-2.6580 | Result='职'
✅ 最终选择输出: 只
--------------------------------------
[上下文以及输入拼音] Context='矩阵的', Pinyin='zhi'
--- Beam Search Top 5 Candidates ---
Rank 1: Score=-0.6708 | Result='值'
Rank 2: Score=-1.7139 | Result='指'
Rank 3: Score=-2.3798 | Result='秩'
Rank 4: Score=-2.4977 | Result='直'
Rank 5: Score=-2.7715 | Result='质'
✅ 最终选择输出: 值
--------------------------------------
[上下文以及输入拼音] Context='', Pinyin='qie'
--- Beam Search Top 5 Candidates ---
Rank 1: Score=-0.4049 | Result='切'
Rank 2: Score=-1.6022 | Result='妾'
Rank 3: Score=-2.9377 | Result='且'
Rank 4: Score=-2.9795 | Result='企鹅'
Rank 5: Score=-4.8103 | Result='惬'
✅ 最终选择输出: 切
--------------------------------------
[上下文以及输入拼音] Context='南极大陆有很多', Pinyin='qie'
--- Beam Search Top 5 Candidates ---
Rank 1: Score=-0.6191 | Result='企鹅'
Rank 2: Score=-1.4296 | Result='且'
Rank 3: Score=-1.9303 | Result='妾'
Rank 4: Score=-3.3165 | Result='切'
Rank 5: Score=-4.4004 | Result='惬'
✅ 最终选择输出: 企鹅
--------------------------------------
