OCR Caffe/Tensorflow/Torch implementation that can be deployed on jetson TX2

I’m a total beginner in deep learning. I’m looking for a OCR program implemented using caffe/Tensorflow/Torch that is specially designed to run on gpu
so that there isn’t too much load on the embedded cpu. I want to train the model on a desktop computer and then deploy the trained model on Jetson TX2
. There are hundreds and even more open-source examples on Github. But I’m not sure which one is more effective. That’s why I’m seeking your suggestions on an open source solution that is robust and able to run on Jetson TX2
. Thanks