questions about using nvprof to profiler caffemodel

I am using caffemodel to classify some images and want to use nvprof to obtain the kernels execution information on the GPU.As we all know, Caffe provides c++ inference and python inference to load caffemodel and inference.
①c++
“nvprof --print-gpu-trace ./build/examples/cpp_classification/classification.bin models/bvlc_alexnet/deploy.prototxt models/bvlc_alexnet/bvlc_alexnet.caffemodel data/ilsvrc12/imagenet_mean.binaryproto data/ilsvrc12/synset_words.txt examples/images/cat”
when I use classification.cpp to inference, nvprof can provide both memory activities and kernels activities.
②python
“nvprof --print-gpu-trace python deep_learnnig_with_python.py”
when I use python inference, nvprof could only provide memory activities but no kernels activities. They all load net model and call net::Forward() to inference , but I don’t know why using nvprof to profiler the latter cannot get the kernels activities. How can I solve it ?