search

med-air / Endo-FM

[MICCAI'23] Foundation Model for Endoscopy Video Analysis via Large-scale Self-supervised Pre-train
Apache License 2.0
146 stars 14 forks source link

GPU requirement #4

Closed weiaicunzai closed 8 months ago

weiaicunzai commented 8 months ago

Hi, thanks for your great work. I have noticed that you have set --nproc_per_node=4 in training script train_clips32k.sh , so could you please tell us if four GPUs with smaller GPU memories (2080ti, Titan V) can be used for training? (without the OOM error)

Kyfafyd commented 8 months ago

halo, thanks for your interest! Four GPUs with smaller memories also can be used for pre-training, you can just decrease the batch_size_per_gpu to fit your GPUs.

weiaicunzai commented 8 months ago

halo, thanks for your interest! Four GPUs with smaller memories also can be used for pre-training, you can just decrease the batch_size_per_gpu to fit your GPUs.

Thank you, I have watched the presentation video of your work on VALSE. And I have noticed that you have used eight 3090 GPUs in your experiment. Could please share with us how many days you have used for pre-training?

Kyfafyd commented 8 months ago

It should take about 1 day for pre-training.

weiaicunzai commented 8 months ago

It should take about 1 day for pre-training.

Thanks.