deberta-v3-large finetuned on a mixture of three sentiment classification datasets
train set
sst2
financial_phrasebank
imdb
test set
sst2
imdb
train args
max_input_length: 256
epoch: 3
optimizer: AdamW
lr: 1e-5
fp16: True
performance
accuracy on mixture of sst2 and imdb test set: 95.3%