showlab / UniVTG

[ICCV2023] UniVTG: Towards Unified Video-Language Temporal Grounding
https://arxiv.org/abs/2307.16715
MIT License
315 stars 28 forks source link

Memory and Time for pretraining #4

Closed wjun0830 closed 1 year ago

wjun0830 commented 1 year ago

Hello Kevin!

Can you provide how much memory space and time is spent for doing PT?

Thanks

QinghongLin commented 1 year ago

Hi @wjun0830 ,

For pretraining, we run on 8GPUs, with less than 24 GB when bsz =32; 10 epoch typically requires 3-4 days; You can flexible decrease the bsz or transformer project dimension for less memory usage and higher efficiency

wjun0830 commented 1 year ago

Thank you!