python279

lib_lightgbm.so: cannot apply additional memory protection after relocation

1

spark: 2.3.1 mmlspark: 0.17 linux: centos6 built gcc-9.1.0, glibc-2.14.1 training data: 5million since i cannot copy & paste the code and log, so attach the captures.  

make it compitable with other torch devices which support fp16 dtype

2

1. specify dtype for specific devices 2. convert the depth to float after model forward