Hello, today’s tutorial is about installing Apache TVM on your Mac OS. Some steps are required for beginners to install it successfully. Friends, Apache TVM is an open-source ML compiler framework for GPUS, CPUs, and deep learning accelerators. The goal of TVM is to enable the deep learning community to optimize and then run computations efficiently on hardware. TVM has a diverse community of hardware vendors, compiler engineers, and ML researchers. This community has built a unified, programmable software stack to enrich the ML technology ecosystem and make it available to everyone. As a result of this collective effort, the performance goes up. We require the following steps to install TVM on MacOS successfully. Please note that Homebrew is required for either Intel or Apple’s M1 chip processors. We need cmake, llvm and Python3.8; please note that currently, Python3.9 have some issues, so I will highly recommend Python3.8
brew install gcc git cmake brew install llvm
If Python3.8 is not installed then either use virtualenv or conda env to install it, otherwise use
brew install firstname.lastname@example.org
By default brew will not link your llvm installation to correct path, for this you need to force link it
brew link llvm --force
git clone --recursive https://github.com/apache/tvm tvm
cd tvm mkdir build cp cmake/config.cmake build
We need to edit edit build/config.cmake file to customize the compilation options, so we can directly add
set(USE_LLVM ON) and let cmake search for a usable version of LLVM. After that we will build and make tvm.
cd build cmake .. make -j4
Once all make is finished successfully, we need to install Python package for this tvm, so that we can use it in Python codes.
export MACOSX_DEPLOYMENT_TARGET=10.9 cd python; python setup.py install --user; cd ..
Below are some useful links about TVM:
 Tutorials: https://tvm.apache.org/docs/tutorials/index.html#autoscheduler-template-free-auto-scheduling
 Benchmark repo: https://github.com/tlc-pack/TLCBench
 OSDI Paper: Ansor : Generating High-Performance Tensor Programs for Deep Learning
 Results on Apple M1 chip: https://medium.com/octoml/on-the-apple-m1-beating-apples-core-ml-4-with-30-model-performance-improvements-9d94af7d1b2d.