Install SparseTIR
Currently we only support build SparseTIR from source code, you’ll need to build the shared library (which is written in C++) first and then install Python bindings. It’s worth noting that SparseTIR is a fork of Apache TVM project, and you don’t need to install Apache TVM to use Sparse TIR.
Pre-requisites
We recommend user to install following packages before compiling SparseTIR shared library:
A recent C++ compiler supporting C++ 14.
CMake 3.18 or higher
LLVM 10 or higher
CUDA Toolkit 11 or higher
Python 3.9 or higher
Install Python Binding
If compilation is successful, the next step is to install SparseTIR binding for Python, you can either install Python package via:
cd python python3 setup.py install
or set environment variable ${PYTHONPATH}:
export SPARSETIR_PATH=$(pwd) export PYTHONPATH=${SPARSETIR_PATH}/python:${PYTHONPATH}