Say if I have data like below
“sample sentence …” “positive”
“sample sentence …” “negative”
if I want to perform binary text classification using T5 on the data above,
when mapping my Target into input ids, is it necessary to append EOS token at the end of Target or it is optional
Specifically, do the two representation below both work for T5 model ?
Target with EOS token: [“id of positive”, “id of EOS token”]
Target without EOS token: [“id of positive”]