brainstorming
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).
install-buildGuides Apache TVM source/PyPI installation, CMake/LLVM builds, import validation, backend probes, and focused test selection.
Independent GitHub snapshot · Automated classification · Original documentation governs
Use this route when the task is to install Apache TVM, configure/build a checkout, validate which libraries Python loads, or choose the smallest native test after a change.
references/build-and-test.md for the
build matrix and focused commands.python/ on
PYTHONPATH; do not use editable TVM/tvm-ffi installs for multi-checkout
development.scripts/check_tvm_install.py before
broader tests. It reports version, loaded libraries, libinfo(), and
backend readiness without changing the checkout.references/troubleshooting.md for
stale imports, missing tvm-ffi, LLVM/linker failures, or optional extras.# From a TVM checkout with submodules initialized.
git submodule update --init --recursive
mkdir -p build
cmake -S . -B build -G Ninja \
-DCMAKE_BUILD_TYPE=RelWithDebInfo \
-DTVM_BUILD_PYTHON_MODULE=ON \
-DUSE_LLVM="llvm-config --ignore-libllvm --link-static" \
-DUSE_RPC=ON -DUSE_CUDA=OFF
cmake --build build --parallel
export PYTHONPATH="$PWD/python${PYTHONPATH:+:$PYTHONPATH}"
export TVM_LIBRARY_PATH="$PWD/build/lib"
python -c 'import tvm; print(tvm.__version__); print(tvm.support.libinfo())'
Adjust USE_CUDA, USE_ROCM, USE_VULKAN, USE_METAL, or optional library
flags only after the matching toolkit, headers, driver, and runtime have been
verified. A successful CMake build does not prove a device backend works.
../relax-compile/SKILL.md.../tirx-kernels/SKILL.md.../s-tir-tuning/SKILL.md.../rpc-deployment/SKILL.md.