docker: add CUDA13 support in dockerfile and update GDRCopy/NVSHMEM for blackwell support (#11517)

Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
This commit is contained in:
ishandhanani
2025-10-27 22:00:54 -07:00
committed by GitHub
co-authored by Baizhou Zhang
parent 813bd6f85c
commit 285a8e6986
7 changed files with 226 additions and 59 deletions
+2 -2
View File
@@ -55,7 +55,7 @@ else
$PIP_CMD install flashinfer-python==0.4.1 --prerelease=allow $PIP_INSTALL_SUFFIX
# Install the main package
$PIP_CMD install -e "python[dev]" --extra-index-url https://download.pytorch.org/whl/${CU_VERSION} $PIP_INSTALL_SUFFIX --upgrade
$PIP_CMD install -e "python" --extra-index-url https://download.pytorch.org/whl/${CU_VERSION} $PIP_INSTALL_SUFFIX --upgrade
fi
# Install router for pd-disagg test
@@ -68,7 +68,7 @@ echo "SGL_KERNEL_VERSION_FROM_KERNEL=${SGL_KERNEL_VERSION_FROM_KERNEL} SGL_KERNE
if [ "${CUSTOM_BUILD_SGL_KERNEL:-}" = "true" ]; then
ls -alh sgl-kernel/dist
$PIP_CMD install sgl-kernel/dist/sgl_kernel-${SGL_KERNEL_VERSION_FROM_KERNEL}-cp310-abi3-manylinux2014_x86_64.whl --force-reinstall $PIP_INSTALL_SUFFIX
$PIP_CMD install sgl-kernel/dist/sgl_kernel-${SGL_KERNEL_VERSION_FROM_KERNEL}+${CU_VERSION}-cp310-abi3-manylinux2014_x86_64.whl --force-reinstall $PIP_INSTALL_SUFFIX
else
$PIP_CMD install sgl-kernel==${SGL_KERNEL_VERSION_FROM_SRT} --force-reinstall $PIP_INSTALL_SUFFIX
fi