Concentrate information about a model

[superchromix]

consider a two-part organization, consisting of a C++ object which stores information about the model, and a CUDA code block which handles the calculations

[jkfindeisen]

Related is the ability to include new models without compiling gpufit (runtime CUDA compilation) #9

[mscipio]

I agree this is an interesting step to consider.

I tried to do a similar thing to what you did here using just PyCuda to make python able to talk with the GPU, avoiding C++ coding. I just made the repo public after submitting a paper using it.

Your LevMar implementation is way beyond mine, but if you are interested I can think of extending the python interface to your library using PyCuda or CuPy as python frontends, and wrap both projects in one. Let me know ...