4 ms·
When building something that I want to run on both CPU and GPU, depending, I’ve found it much easier to use PyTorch than some combination of NumPy and CuPy. I d
by meisel 2y ago
When building something that I want to run on both CPU and GPU, depending, I’ve found it much easier to use PyTorch than some combination of NumPy and CuPy. I don’t have to fiddle around with some global replacing of numpy.* with cupy.*, and PyTorch has very nearly all the functions that those libraries have.
- setopt 2y agoInteresting. Any links to examples or docs on how to use PyTorch as a general linear algebra library for this purpose? Like a “SciPy to PyTorch” transition guide if I want to do the same?