Why not fine-tune on TOEIC dataset?

Open jihun-hong opened this issue 6 years ago • 1 comments

Is there a reason that you didn't fine-tune BERT for solving TOEIC problems? Is it because the dataset for TOEIC problem solving is too small for fine-tuning?

Edit: You seem to have more than 7,000 examples of TOEIC blank-filling problems. This seems enough to fine-tune BERT for this task. Was there a reason that you didn't try that?

Aug 23 '19 07:08 jihun-hong

No I just only want to see a result with contextual representation of pretrained BERT.

Aug 23 '19 08:08 graykode