'How BERT tokenizer works for subwords?
For WordPiece tokenizer, we split the tokens like playing to play and ##ing. And we get embedding for play and ##ing separately as the representation of playing. Now I want only one representation of playing. To do this, I can do elementwise multiplication or addition of play and ##ing. But which is correct or is there any way to do this?
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|