Closed yhshu closed 3 years ago
Exactly~
I haven't actually experimented to the stage of generating results, so there's a question.
The transformer tokenizer, i.e., BART tokenizer here, generally separates symbols like <
>
from words.
Is there any post-process for seq2seq results to eliminate the side effect, or deal with the possible ill-format SPARQL or Program?
For SPARQL,there is post-process as shown in https://github.com/shijx12/KQAPro_Baselines/blob/2c5049900ba11947daf32cffcdfbdb4e985f1108/Bart_SPARQL/predict.py#L90.
I assume that
<b>
splits functions, and<c>
splits arguments of functions, is it true?Thanks for your reply.