ZixuanKe/PyContinual

Running CTR and B-CL takes a long time

rahmahkn opened this issue · 1 comments

rahmahkn commented

Hi! Thank you for your excellent work for PyContinual. I wonder why it takes a long time (3-7 days) to run CTR and B-CL with the bert_adapter backbone. I am using my dataset with 17 tasks of sentiment analysis and emotion classification. I would like to know your GPU specification or if you have experienced the same thing. Thanks!

ZixuanKe commented

Hi,

Thank you so much for your interest in our work.

It could be slow if you use the main branch as it lacks support for multi-node and fp16. You may try the dev branch to achieve faster speeds.

  • 👍1