A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 16:00 UTC
Patch DeepEP when building docker image with CUDA 13 (#29154)
Signed-off-by: Qidong Su <soodoshll@gmail.com>
Qidong Su committed
Nov 22, 2025 at 18:25 UTC
4587063267d2751ac183c16421aeda3f335cee39
1 file changed
+11
tools/ep_kernels/install_python_libraries.sh
+11
@@ -96,6 +96,13 @@ clone_repo() {
96
fi
97
}
98
99
+deepep_cuda13_patch() {
100
+ cuda_version_major=$(${CUDA_HOME}/bin/nvcc --version | egrep -o "release [0-9]+" | cut -d ' ' -f 2)
101
+ if [ ${cuda_version_major} -ge 13 ]; then
102
+ sed -i "s|f'{nvshmem_dir}/include']|f'{nvshmem_dir}/include', '${CUDA_HOME}/include/cccl']|" "setup.py"
103
+ fi
104
+}
105
+
106
do_build() {
107
local repo=$1
108
local name=$2
@@ -107,6 +114,10 @@ do_build() {
114
clone_repo "$repo" "$name" "$key" "$commit"
115
cd "$name"
116
117
+ if [ "$name" == "DeepEP" ]; then
118
+ deepep_cuda13_patch
119
+ fi
120
+
121
if [ "$MODE" = "install" ]; then
122
echo "Installing $name into environment"
123
eval "$extra_env" uv pip install --no-build-isolation -vvv .