search

Koorye / DePT

[CVPR 2024] Offical implemention of the paper "DePT: Decoupled Prompt Tuning"
GNU General Public License v2.0
65 stars 2 forks source link

What GPUs are your experiments running on, and how many are needed? #2

Closed negan7879 closed 2 months ago

Koorye commented 3 months ago

Thank you for your interest in our work. We run our experminents on a single V100 GPU.

negan7879 commented 3 months ago

Thank you for your interest in our work. We run our experminents on a single V100 GPU.

Thanks for the reply, how long does it take to execute on a single V100?

Koorye commented 3 months ago

Thank you for your interest in our work. We run our experminents on a single V100 GPU.

Thanks for the reply, how long does it take to execute on a single V100?

For the base-to-new generalization task, our method takes approximately 4 hours to run on all datasets with a random seed. For cross-dataset transfer tasks, this would take approximately 8 hours.

The above results are based on the CoOp method, with varying training times for each method. Our method does not increase additional training time.