software-engineering/spark-vllm-docker
2
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

Disabled Triton build for now

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-01-31 00:10:52 -08:00
parent 7c61b4057c
commit 9691eed1b0
1 changed files with 24 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
Dockerfile
View File

@@ -64,29 +64,31 @@ ENV TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
FROM base AS builder FROM base AS builder
# Triton Build # ======= Triton Build ==========
# Initial Triton repo clone (cached forever) # # Initial Triton repo clone (cached forever)
RUN git clone https://github.com/triton-lang/triton.git # RUN git clone https://github.com/triton-lang/triton.git
# We expect TRITON_REF to be passed from the command line to break the cache # # We expect TRITON_REF to be passed from the command line to break the cache
# Set to v3.5.1 tag by default # # Set to v3.5.1 tag by default
ARG TRITON_REF=v3.5.1 # ARG TRITON_REF=v3.5.1
WORKDIR $VLLM_BASE_DIR/triton # WORKDIR $VLLM_BASE_DIR/triton
# This only runs if TRITON_REF differs from the last build # # This only runs if TRITON_REF differs from the last build
RUN --mount=type=cache,id=ccache,target=/root/.ccache \ # RUN --mount=type=cache,id=ccache,target=/root/.ccache \
--mount=type=cache,id=uv-cache,target=/root/.cache/uv \ # --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
git fetch origin && \ # git fetch origin && \
git checkout ${TRITON_REF} && \ # git checkout ${TRITON_REF} && \
git submodule sync && \ # git submodule sync && \
git submodule update --init --recursive && \ # git submodule update --init --recursive && \
uv pip install -r python/requirements.txt && \ # uv pip install -r python/requirements.txt && \
mkdir -p /workspace/wheels && \ # mkdir -p /workspace/wheels && \
rm -rf .git && \ # rm -rf .git && \
uv build --no-build-isolation --wheel --out-dir=/workspace/wheels -v . && \ # uv build --no-build-isolation --wheel --out-dir=/workspace/wheels -v . && \
uv build --no-build-isolation --wheel --no-index --out-dir=/workspace/wheels python/triton_kernels # uv build --no-build-isolation --wheel --no-index --out-dir=/workspace/wheels python/triton_kernels
# ======= FlashInfer Build ==========
ENV FLASHINFER_CUDA_ARCH_LIST="12.1f" ENV FLASHINFER_CUDA_ARCH_LIST="12.1f"
WORKDIR $VLLM_BASE_DIR WORKDIR $VLLM_BASE_DIR
@@ -279,8 +281,9 @@ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
# Cleanup # Cleanup
# Remove triton-kernels as they are not compatible with this vLLM version yet # If not compiling Triton
# RUN uv pip uninstall triton-kernels # remove triton-kernels as they are not compatible with this vLLM version yet
RUN uv pip uninstall triton-kernels
# Keeping it here for reference - this won't work as is without squashing layers # Keeping it here for reference - this won't work as is without squashing layers
# RUN uv pip uninstall absl-py apex argon2-cffi \ # RUN uv pip uninstall absl-py apex argon2-cffi \