search

wasiahmad / PLBART

Official code of our work, Unified Pre-training for Program Understanding and Generation [NAACL 2021].
https://arxiv.org/abs/2103.06333
MIT License
186 stars 35 forks source link

Time required for fine-tuning #23

Closed wangdeze18 closed 2 years ago

wangdeze18 commented 2 years ago

Hi, thanks for your great work! When I finetuned downstream tasks on my NVIDIA 3090, I roughly calculated that 100k fine-tuning steps would take 2 days. This does not seem very acceptable. （Especially, CodeBERT takes only a few hours for fine-tuneing for the same task and dataset.）

Any suggestions? Thanks！

wasiahmad commented 2 years ago

Not enough information was provided. Which tasks you are referring to? All the checkpoints are shared along with the log files, and you can check how much time it took for us to finetune the model. In my opinion, the runtime should be faster since we use Fairseq. The comparison with CodeBERT doesn't make sense to me.