I am sick and tired of python, and there's nothing comfortable about it, but without CUDA or seriously tuned opencl backend I'm not going to waste my time training models at 1/50th the output
a
Amir Gur
09/26/2019, 8:01 PM
Yes, of course that if things are not fully baked you'd stay python despite not being comfortable w/ it.
Cuda allows other languages. Is a kotlin (or kotlin-native) implementation needed? There are a few kotlin/cuda projects online. Or can jcuda help?
All OpenCLs are based on LLVM/CLang, guess we could use kotlin-native.
j
jimn
10/02/2019, 8:04 AM
i would be OK with a tensorflow wrapper that came with a pandas expression tool. cuda has several prebaked NN library enhancements that sort of intermingle with tensorflow evolution
jimn
10/02/2019, 8:05 AM
keras is the horizon focal point for DL api's actually. s/tensorflow/keras