SuperGen icon indicating copy to clipboard operation
SuperGen copied to clipboard

Could you please offer a script for gpt2 to generate the training data?

Open Luobots opened this issue 2 years ago • 1 comments

I made a gpt2 script to generate the training data following your paper, but it did badly in qnli task, different to CTRL, the filtering rule is not suitable for gpt2, but I can't handle the rule myself. Thank you.

Luobots avatar Nov 02 '22 17:11 Luobots

Hi,

We have only used GPT-2 for generation on MNLI and SST-2 tasks, but the code adaptation for other tasks should be straightforward. You do need to change a few places like removing the control codes used in CTRL and how the termination of a generated sequence is determined -- this can be done by using certain punctuation marks based on the generation patterns of GPT-2 on the target task.

Best, Yu

yumeng5 avatar Nov 03 '22 05:11 yumeng5