Using GPU in custom python layer

[Sumit Maharjan]

I have been trying to implement some custom loss functions in caffe by defining them in python as implemented in these examples.

https://github.com/BVLC/caffe/blob/master/python/caffe/test/test_python_layer.py

How do you make caffe perform calculation in GPU in these type of loss layer?
Do I need to implement calculations using other libraries cupy, numba etc.

While there is explanation on implementing forward_gpu and backward_gpu for the c++ implementation, I couldn't find any other information regarding the python implemented.

Any solution or suggestion is much appreciated.

[OCR]

Same question here. Does the python layer automatically use GPU or something else should be done?

[OCR]