Configure multiple GPU for training in Python

[Hongmin Wang]

Caffe supports multi-GPU training by setting the "-gpu" argument with multiple device IDs in the command line interface.

Does anyone know how to specify multiple GPUs in Python for training? (e.g.caffe.set_device(GPU_ID)) 

i.e. instead of writing solver and network descriptions into .prototxt files, training with command line, and then testing with Python, can we do all these by just running one python script, for training on multiple GPUs?

Thank you!

[Jeremy Rutman]

I second this question