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 14, 05:30 UTC
[Kernel] Update DeepGEMM to latest commit (#23915)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Jee Jee Li committed
Sep 1, 2025 at 17:38 UTC
dc1a53186d8dbb3b450174a38a10c474b55f22e9
2 files changed
+3
-4
docker/Dockerfile
+2
-3
@@ -432,11 +432,10 @@ RUN --mount=type=cache,target=/root/.cache/uv \
432
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
433
434
# Install DeepGEMM from source
435
-ARG DEEPGEMM_GIT_REF="7b6b5563b9d4c1ae07ffbce7f78ad3ac9204827c"
435
+ARG DEEPGEMM_GIT_REF
436
COPY tools/install_deepgemm.sh /tmp/install_deepgemm.sh
437
RUN --mount=type=cache,target=/root/.cache/uv \
438
- VLLM_DOCKER_BUILD_CONTEXT=1 /tmp/install_deepgemm.sh --cuda-version "${CUDA_VERSION}" --ref "${DEEPGEMM_GIT_REF}" \
439
- && rm /tmp/install_deepgemm.sh
438
+ VLLM_DOCKER_BUILD_CONTEXT=1 /tmp/install_deepgemm.sh --cuda-version "${CUDA_VERSION}" ${DEEPGEMM_GIT_REF:+--ref "$DEEPGEMM_GIT_REF"}
439
440
# Install EP kernels(pplx-kernels and DeepEP), NixL
441
COPY tools/ep_kernels/install_python_libraries.sh install_python_libraries.sh
tools/install_deepgemm.sh
+1
-1
@@ -6,7 +6,7 @@ set -e
6
7
# Default values
8
DEEPGEMM_GIT_REPO="https://github.com/deepseek-ai/DeepGEMM.git"
9
-DEEPGEMM_GIT_REF="7b6b5563b9d4c1ae07ffbce7f78ad3ac9204827c"
9
+DEEPGEMM_GIT_REF="ea9c5d9270226c5dd7a577c212e9ea385f6ef048"
10
11
# Parse command line arguments
12
while [[ $# -gt 0 ]]; do