biondizzle/grace-gpu-containers
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use PyPI triton wheel instead of building (QEMU segfaults)

Browse Source 
Triton 3.6.0 has official aarch64 wheel on PyPI.
Building triton from source causes segfaults under QEMU emulation.
This commit is contained in:
biondizzle
2026-04-02 23:58:20 +00:00
parent e5445512aa
commit 436214bb72
1 changed files with 3 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/Dockerfile
View File

@@ -64,16 +64,10 @@ RUN mkdir /wheels
# Pin setuptools to <81 for LMCache compatibility (needs >=77.0.3,<81.0.0) # Pin setuptools to <81 for LMCache compatibility (needs >=77.0.3,<81.0.0)
RUN uv pip install -U build cmake ninja pybind11 "setuptools>=77.0.3,<81.0.0" wheel RUN uv pip install -U build cmake ninja pybind11 "setuptools>=77.0.3,<81.0.0" wheel
# Use PyPI triton wheel instead of building (QEMU segfaults during triton build)
FROM build-base AS build-triton FROM build-base AS build-triton
ARG TRITON_REF=release/3.5.x RUN mkdir -p /wheels && \
ARG TRITON_BUILD_SUFFIX=+cu130 pip download triton==3.6.0 --platform manylinux_2_27_aarch64 --only-binary=:all: --no-deps -d /wheels
ENV TRITON_WHEEL_VERSION_SUFFIX=${TRITON_BUILD_SUFFIX:-}
RUN git clone https://github.com/triton-lang/triton.git
RUN cd triton && \
git checkout ${TRITON_REF} && \
git submodule sync && \
git submodule update --init --recursive -j 8 && \
uv build --wheel --no-build-isolation -o /wheels
FROM build-base AS build-xformers FROM build-base AS build-xformers
#ARG XFORMERS_REF=v0.0.32.post2 #ARG XFORMERS_REF=v0.0.32.post2