allenai / container

MIT License
56 stars 9 forks source link

About the training configures for Container-Base model #1

Closed ChongjianGE closed 2 years ago

ChongjianGE commented 2 years ago

Hi,

Thanks for the interesting work and the provided code for container_v1_light. I would like to ask how many images per batch are used in training the container_base model, since it may simply cause the OOM error when training the container_base model.

gaopengcuhk commented 2 years ago

We didn't release container_base. Where do you find container_base?

gaopengcuhk commented 2 years ago

container_base is trained with 8 * 48G GPU Memory.

gaopengcuhk commented 2 years ago

container_base is trained with 8 * 48G GPU Memory, 128 images per GPU.

gaopengcuhk commented 2 years ago

If you need container-base configure, I can share it.

ChongjianGE commented 2 years ago

Hi @gaopengcuhk , Thanks for the reply on the training details. Speaking of the container_base, I simply change part of the code following the descriptions in the paper. It will be great if you can share the container_base configuration file with me (via rhettgee@connect.hku.hk if it's convenient for you). Thanks again.