Training with multiple gpus on pycaffe

[André Teixeira Lopes]

Hi,

As far as know, caffe supports multiple GPUs on training.

After some research, unfortunatelly, I didn't find a way to set multiple GPUs in the pycaffe interface for training. 

I know we can use multiple GPU over the command line

caffe train -solver examples/mnist/lenet_solver.prototxt -gpu 0,1

But my data comes from a memory_data_layer, so, this isn't a feasible solution for me. 

Does someone know how to enable multiple GPUs on the pycaffe API, in the training phase?

Thaks a lot!