Hi, I’d like to depend on TVM’s Python interface in my repo, so I need an easy way to pull in the TVM project during CI.
Since the setup instructions are a little involved and take a while, I see a few ways to do this:
- Write a shell script that installs TVM via the TVM setup instructions.
- Build a Docker container that has TVM pre-built into it. This would make CI faster because I wouldn’t have to re-build the TVM C++ code.
Both of these ideas are doable but don’t feel that clean, and may require extra work in the future if the setup instructions change. Has anyone found a better way to do this, or does TVM have a recommended way to do this?