Re:Why ML Needs a New Programming Language
by chickenzzzzu on Saturday September 06, 2025 @12:46AM
(#45146697)
Not only is their argument circular but it is wrong. There is no need to use 50 million lines of Python, Pytorch, Numpy, Linux, Cmake, CUDA, and god knows how many other layers of madness to do inference.
It is literally on the order of tens of thousands of lines of code, instead of tens of millions, to do Vulkan ML, especially if you strip out the parts of the kernel you don't need.
It is literally on the order of tens of thousands of lines of code, instead of tens of millions, to do Vulkan ML, especially if you strip out the parts of the kernel you don't need.