I already installed tvm on my own pc with llvm and vs2017, and now my python project can run.
But now I want to release my project to other win10 computers with python environment, must I compile all tvm project with llvm and vs ? (It will take nuch long time and harddisk space).
To just release your project (save your built model to a dll), I don’t think you will need those dependencies. However, to run the full TVM pipeline and build your model on other machines you definitely will.
I found that for GPU, I need CUDA, a subset of a Visual Studio installation (NVCC depends on it), and the built LLVM binary. For CPU, I found that I only needed the built LLVM binary.