This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm-tirx.git
The following commit(s) were added to refs/heads/main by this push:
new b05b7ae68 Require tvm-ffi 0.1.16 and drop the sidecar tvm-ffi install
(#240)
b05b7ae68 is described below
commit b05b7ae68494bdb916a269a6c4cc799c4e9d8e55
Author: Tianqi Chen <[email protected]>
AuthorDate: Wed Nov 11 10:06:46 2026 -0500
Require tvm-ffi 0.1.16 and drop the sidecar tvm-ffi install (#240)
tvm-ffi 0.1.16 is released, on the same commit as `v0.1.16-rc4`, which
the submodule already pins. This requires `apache-tvm-ffi>=0.1.16` in
`pyproject.toml` in place of `>=0.1.16rc4`.
The CUDA runtime sidecar build scripts no longer install
`apache-tvm-ffi`. The sidecar CMake build takes tvm-ffi from the
submodule (`add_subdirectory(3rdparty/tvm-ffi)`) and never uses the
Python package, so `pyproject.toml` is the only place that names a
tvm-ffi version.
---------
Signed-off-by: tqchen <[email protected]>
---
ci/wheel/manylinux_build_libtvm_runtime_cuda.sh | 1 -
ci/wheel/windows_build_libtvm_runtime_cuda.bat | 2 +-
pyproject.toml | 2 +-
3 files changed, 2 insertions(+), 3 deletions(-)
diff --git a/ci/wheel/manylinux_build_libtvm_runtime_cuda.sh
b/ci/wheel/manylinux_build_libtvm_runtime_cuda.sh
index 385e06e02..f2573586a 100755
--- a/ci/wheel/manylinux_build_libtvm_runtime_cuda.sh
+++ b/ci/wheel/manylinux_build_libtvm_runtime_cuda.sh
@@ -39,7 +39,6 @@ dnf -y install cuda-driver-devel-13-1
# generator (no Ninja), so no build tools are installed here. Put the bundled
# CPython and CUDA toolchain on PATH for the CMake configure and nvcc.
export PATH="/opt/python/cp310-cp310/bin:/usr/local/cuda/bin:${PATH}"
-python -m pip install "apache-tvm-ffi>=0.1.16rc4"
nvcc --version
rm -rf "${build_dir}"
diff --git a/ci/wheel/windows_build_libtvm_runtime_cuda.bat
b/ci/wheel/windows_build_libtvm_runtime_cuda.bat
index 4e4a7fb7b..a1d507528 100644
--- a/ci/wheel/windows_build_libtvm_runtime_cuda.bat
+++ b/ci/wheel/windows_build_libtvm_runtime_cuda.bat
@@ -86,7 +86,7 @@ set "cuda_root=%CD%"
popd
set "CUDA_PATH=%cuda_root%"
-python -m pip install -U pip cmake ninja "apache-tvm-ffi>=0.1.16rc4" || exit
/b 1
+python -m pip install -U pip cmake ninja || exit /b 1
"%nvcc_exe%" --version || exit /b 1
rem nvcc needs the MSVC host compiler (cl.exe), so locate VS via vswhere and
run the
diff --git a/pyproject.toml b/pyproject.toml
index 726eaaac9..35c185188 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -38,7 +38,7 @@ classifiers = [
"Topic :: Software Development :: Libraries :: Python Modules",
]
dependencies = [
- "apache-tvm-ffi>=0.1.16rc4",
+ "apache-tvm-ffi>=0.1.16",
"apache-tvm-ffi-orcjit>=0.1.3",
"ml_dtypes",
"numpy",