I have been trying to use NVCaffe’s FP16 support through DIGITS, however despite setting
default_forward_type: FLOAT16 default_backward_type: FLOAT16 default_forward_math: FLOAT16 default_backward_math: FLOAT16
the caffe_output.log states that is is using regular FLOAT instead.
I0706 15:49:26.557221 824 net.cpp:109] Using FLOAT as default forward math type I0706 15:49:26.557232 824 net.cpp:115] Using FLOAT as default backward math type I0706 15:49:26.557240 824 layer_factory.hpp:172] Creating layer 'train-data' of type 'Data' I0706 15:49:26.557252 824 layer_factory.hpp:184] Layer's types are Ftype:FLOAT Btype:FLOAT Fmath:FLOAT Bmath:FLOAT
Is this supported in DIGITS at all? and if so what am I missing