Every package is a module, and Julia is well-known to not have monorepos and instead split packages into small modules. So I don't quite see how this follows: t

Every package is a module, and Julia is well-known to not have monorepos and instead split packages into small modules. So I don't quite see how this follows: there's no PyTorch, instead there's Flux, NNLib, NNLibCUDA, NNLibAMDGPU, CUDA, GPUCompiler, KernalAbstractions, ... (it keeps going), all of which are documented packages in their own modules that make up the ML stack.