how to set multi-GPUs

[Ai-is-light]

when I train the network, I got the information as following: (I build the caffe-training environment under the https://github.com/yjxiong/temporal-segment-networks) You are running caffe compiled with MPI support. Now it's running in non-parallel model what's meaning?

What's more, I can't use the multi-gpus to train my model.

Looking forward to any advice.

[Ai-is-light]

@mzolfaghari would you mind supporting your caffe docker images? Thanks

[mzolfaghari]

@Ai-is-light

• We compiled our Caffe without MPI suppert! Instead we increased "iter_size" in the solver file.

• Right now I'm busy but we'll provide our Caffe docker images in the near future. Currently, you can check the following link for the compile setting: https://github.com/yjxiong/caffe/tree/3D

• The PyTorch version will be available in the coming weeks.

[Ai-is-light]

@mzolfaghari Thanks, I also set the 'iter_size' for increasing the batch size.

[sophia-wright-blue]

Hello @mzolfaghari ,

Thank you for making the code available - I have a follow-up on your PyTorch comment - is there a scheduled release date for the PyTorch version?

Thanks,

[mzolfaghari]

Hi @sophia-wright-blue

The code will be available next week.

[sophia-wright-blue]

That's awesome - looking forward to learning more about your model.