guided_summarization about the wikihow train script

Hi, I have comparable results to those in paper on CNNDM dataset. However, when I run the same train.sh scripts for wikihow dataset, it can't generate right sentences. So I'd appreciate it if you could release the train script for wikihow, namely, the lr(s), warm_up steps...... Thanks!

Jun 17 '21 02:06 DavidJiang9

Thanks! I think the hyper-parameters for WikiHow and CNNDM are the same. Just to confirm, could you get comparable performance of BART and MatchSum on WikiHow?

Jun 17 '21 02:06 zdou0830

I haven't tried it yet(I use bertabs + bertext). What makes me confused is that the performance drop sharply (sudden decrease in accuracy and increase in ppl from tensorboard) at step around 2K. Does changing models helps? BTW, I use like 50% of the original WikiHow dataset not 100%. Hope to get your advice.

Jun 17 '21 03:06 DavidJiang9

Hi, I didn't look at the accuracy or ppl during training. I think if you can reproduce the baseline performance and your guidance extraction process is correct, you could get a similar performance with ours using the same hyper-parameters with CNNDM.

Jun 17 '21 04:06 zdou0830

OK，thanks a lot. I will try the baseline on WikiHow.

Jun 17 '21 05:06 DavidJiang9