Distributed Training with PyTorch on Multi-GPU Setup

[ballin0105]

Hi there,

I've been following your work with great interest and appreciate all the effort you've put into it. I encountered an issue when running your code on a remote server with 8 V-100 GPUs under PyTorch. After switching the launcher to PyTorch, I ran into a address already in use error that seems to prevent multi-GPU utilization, restricting the process to a single GPU.

Is there any chance that a distributed training update compatible with PyTorch might be on the horizon? It would greatly benefit those of us working with similar hardware configurations.

Thanks for your continued contributions to the field!

[Jingkang50]

Thank you for your interest in our work! The distributed training framework is based on MMDet, maybe you can look for some solutions at the MMDet repo or community? If you still cannot solve it, please lemme know.
Thanks!

[ballin0105]

Thank you for your interest in our work! The distributed training framework is based on MMDet, maybe you can look for some solutions at the MMDet repo or community? If you still cannot solve it, please lemme know. Thanks!

Thank you so much for your helpful response! :)

[leaozhun]

Hello,have you resolved the issue with PyTorch distributed training?