OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 00:20:18 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[vllm hash update] update the pinned vllm hash (#164628)

Browse Source 
This PR is auto-generated nightly by [this action](https://github.com/pytorch/pytorch/blob/main/.github/workflows/nightly.yml).
Update the pinned vllm hash.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/164628
Approved by: https://github.com/pytorchbot

Co-authored-by: Huy Do <huydhn@gmail.com>
This commit is contained in:
PyTorch UpdateBot 2025-10-12 18:26:07 +00:00 committed by PyTorch MergeBot
parent 2beead7523
commit a2601630cd
5 changed files with 92 additions and 193 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.ci/lumen_cli/cli/lib/core/vllm/lib.py
View File

@ -143,7 +143,7 @@ def sample_vllm_test_library():
"pytest -v -s compile/test_decorator.py", "pytest -v -s compile/test_decorator.py",
], ],
}, },
"vllm_languagde_model_test_extended_generation_28_failure_test": { "vllm_language_model_test_extended_generation_28_failure_test": {
"title": "Language Models Test (Extended Generation) 2.8 release failure", "title": "Language Models Test (Extended Generation) 2.8 release failure",
"id": "vllm_languagde_model_test_extended_generation_28_failure_test", "id": "vllm_languagde_model_test_extended_generation_28_failure_test",
"package_install": [ "package_install": [

2
.ci/lumen_cli/cli/lib/core/vllm/vllm_build.py
View File

@ -63,7 +63,7 @@ class VllmBuildParameters:
# DOCKERFILE_PATH: path to Dockerfile used when use_local_dockerfile is True" # DOCKERFILE_PATH: path to Dockerfile used when use_local_dockerfile is True"
use_local_dockerfile: bool = env_bool_field("USE_LOCAL_DOCKERFILE", True) use_local_dockerfile: bool = env_bool_field("USE_LOCAL_DOCKERFILE", True)
dockerfile_path: Path = env_path_field( dockerfile_path: Path = env_path_field(
"DOCKERFILE_PATH", ".github/ci_configs/vllm/Dockerfile.tmp_vllm" "DOCKERFILE_PATH", ".github/ci_configs/vllm/Dockerfile"
) )
# the cleaning script to remove torch dependencies from pip # the cleaning script to remove torch dependencies from pip

2
.github/ci_commit_pins/vllm.txt vendored
View File

@ -1 +1 @@
0ad9951c416d33c5da4f7a504fb162cbe62386f5 e5192819208c4d68194844b7dfafbc00020d0dea

275
.github/ci_configs/vllm/Dockerfile.tmp_vllm → .github/ci_configs/vllm/Dockerfile vendored
View File

@ -1,59 +1,71 @@
# TODO(elainwy): remove this file after the torch nightly dockerfile is in sync in vllm repo
# The vLLM Dockerfile is used to construct vLLM image against torch nightly and torch main that can be directly used for testing
ARG CUDA_VERSION=12.8.1 ARG CUDA_VERSION=12.8.1
ARG PYTHON_VERSION=3.12 ARG PYTHON_VERSION=3.12
# BUILD_BASE_IMAGE: used to setup python build xformers, and vllm wheels, It can be replaced with a different base image from local machine, # BUILD_BASE_IMAGE: used to setup python build xformers, and vllm wheels, It can be replaced with a different base image from local machine,
# by default, it uses the torch-nightly-base stage from this docker image # by default, it uses the torch-nightly-base stage from this docker image
ARG BUILD_BASE_IMAGE=torch-nightly-base ARG BUILD_BASE_IMAGE=torch-nightly-base
# FINAL_BASE_IMAGE: used to set up vllm-instaled environment and build flashinfer,
# by default, it uses devel-ubuntu22.04 official image.
ARG FINAL_BASE_IMAGE=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04 ARG FINAL_BASE_IMAGE=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04
# The logic is copied from https://github.com/vllm-project/vllm/blob/main/docker/Dockerfile # The logic is copied from https://github.com/vllm-project/vllm/blob/main/docker/Dockerfile
ARG GET_PIP_URL="https://bootstrap.pypa.io/get-pip.py" ARG GET_PIP_URL="https://bootstrap.pypa.io/get-pip.py"
#################### TORCH NIGHTLY BASE IMAGE #################### #################### TORCH NIGHTLY BASE IMAGE ####################
# A base image for building vLLM with devel ubuntu 22.04, this is mainly used to build vllm in vllm builtkite ci
FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04 as torch-nightly-base FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04 as torch-nightly-base
ARG CUDA_VERSION ARG CUDA_VERSION
ARG PYTHON_VERSION ARG PYTHON_VERSION
ARG GET_PIP_URL ARG GET_PIP_URL
# Install Python and other dependencies # Install system dependencies and uv, then create Python virtual environment
RUN apt-get update -y \ RUN apt-get update -y \
&& apt-get install -y ccache software-properties-common git curl wget sudo vim \ && apt-get install -y ccache software-properties-common git curl sudo vim python3-pip \
&& add-apt-repository -y ppa:deadsnakes/ppa \ && curl -LsSf https://astral.sh/uv/install.sh | sh \
&& apt-get update -y \ && $HOME/.local/bin/uv venv /opt/venv --python ${PYTHON_VERSION} \
&& apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \ && rm -f /usr/bin/python3 /usr/bin/python3-config /usr/bin/pip \
&& update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \ && ln -s /opt/venv/bin/python3 /usr/bin/python3 \
&& update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \ && ln -s /opt/venv/bin/python3-config /usr/bin/python3-config \
&& ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \ && ln -s /opt/venv/bin/pip /usr/bin/pip \
&& curl -sS ${GET_PIP_URL} | python${PYTHON_VERSION} \
&& python3 --version && python3 -m pip --version && python3 --version && python3 -m pip --version
# Upgrade to GCC 10 to avoid https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92519 # Upgrade to GCC 10 to avoid https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92519
# as it was causing spam when compiling the CUTLASS kernels # as it was causing spam when compiling the CUTLASS kernels
# Ensure gcc >= 10 to avoid CUTLASS issues (bug 92519) RUN apt-get install -y gcc-10 g++-10
RUN current_gcc_version=$(gcc -dumpversion | cut -f1 -d.) && \ RUN update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-10 110 --slave /usr/bin/g++ g++ /usr/bin/g++-10
if command -v apt-get >/dev/null; then \ RUN <<EOF
if [ "$current_gcc_version" -lt 10 ]; then \ gcc --version
echo "GCC version is $current_gcc_version, installing gcc-10..."; \ EOF
apt-get update \
&& apt-get install -y gcc-10 g++-10 \
&& update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-10 100 \
&& update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-10 100; \
else \
echo "GCC version is $current_gcc_version, no need to install gcc-10."; \
fi \
fi \
&& gcc --version && g++ --version
# install uv for faster pip installs # Install uv for faster pip installs
RUN --mount=type=cache,target=/root/.cache/uv \
python3 -m pip install uv==0.8.4
ENV UV_HTTP_TIMEOUT=500
ENV UV_INDEX_STRATEGY="unsafe-best-match"
# Use copy mode to avoid hardlink failures with Docker cache mounts
ENV UV_LINK_MODE=copy
#################### TORCH NIGHTLY BASE IMAGE ####################
#################### BASE BUILD IMAGE ####################
FROM ${BUILD_BASE_IMAGE} AS base
USER root
ARG CUDA_VERSION
ARG PYTHON_VERSION
# Only work with PyTorch manylinux builder
ENV PATH="/opt/python/cp312-cp312/bin:${PATH}"
# Install some system dependencies and double check python version
RUN if command -v apt-get >/dev/null; then \
apt-get update -y \
&& apt-get install -y ccache software-properties-common git wget sudo vim; \
else \
dnf install -y git wget sudo; \
fi \
&& python3 --version && python3 -m pip --version
# Install uv for faster pip installs if not existed
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
python3 -m pip install uv==0.8.4 python3 -m pip install uv==0.8.4
@ -62,51 +74,17 @@ ENV UV_INDEX_STRATEGY="unsafe-best-match"
# Use copy mode to avoid hardlink failures with Docker cache mounts # Use copy mode to avoid hardlink failures with Docker cache mounts
ENV UV_LINK_MODE=copy ENV UV_LINK_MODE=copy
#################### TORCH NIGHTLY BASE IMAGE ####################
#################### BASE BUILD IMAGE ####################
# A base image for building vLLM with torch nightly or torch wheels
# prepare basic build environment
FROM ${BUILD_BASE_IMAGE} AS base
USER root
ARG CUDA_VERSION
ARG PYTHON_VERSION
# TODO (huydhn): Only work with PyTorch manylinux builder
ENV PATH="/opt/python/cp312-cp312/bin:${PATH}"
# Install some system dependencies and double check python version
RUN if command -v apt-get >/dev/null; then \
apt-get update -y \
&& apt-get install -y ccache software-properties-common git curl wget sudo vim; \
else \
dnf install -y git curl wget sudo; \
fi \
&& python3 --version && python3 -m pip --version
# Install uv for faster pip installs if not existed
RUN --mount=type=cache,target=/root/.cache/uv \
if ! python3 -m uv --version >/dev/null 2>&1; then \
python3 -m pip install uv==0.8.4; \
fi
ENV UV_HTTP_TIMEOUT=500
ENV UV_INDEX_STRATEGY="unsafe-best-match"
# Use copy mode to avoid hardlink failures with Docker cache mounts
ENV UV_LINK_MODE=copy
WORKDIR /workspace WORKDIR /workspace
# install build and runtime dependencies # Install build and runtime dependencies
COPY requirements/common.txt requirements/common.txt COPY requirements/common.txt requirements/common.txt
COPY use_existing_torch.py use_existing_torch.py COPY use_existing_torch.py use_existing_torch.py
COPY pyproject.toml pyproject.toml COPY pyproject.toml pyproject.toml
# install build and runtime dependencies without stable torch version # Install build and runtime dependencies without stable torch version
RUN python3 use_existing_torch.py RUN python3 use_existing_torch.py
# default mount file as placeholder, this just avoid the mount error # Default mount file as placeholder, this just avoid the mount error
# change to a different vllm folder if this does not exist anymore # change to a different vllm folder if this does not exist anymore
ARG TORCH_WHEELS_PATH="./requirements" ARG TORCH_WHEELS_PATH="./requirements"
ARG PINNED_TORCH_VERSION ARG PINNED_TORCH_VERSION
@ -138,56 +116,36 @@ RUN --mount=type=cache,target=/root/.cache/uv \
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system -r requirements/common.txt uv pip install --system -r requirements/common.txt
# Must put before installing xformers, so it can install the correct version of xfomrers.
ARG xformers_cuda_arch_list='7.5;8.0+PTX;9.0a'
ENV TORCH_CUDA_ARCH_LIST=${xformers_cuda_arch_list}
ARG max_jobs=16 ARG max_jobs=16
ENV MAX_JOBS=${max_jobs} ENV MAX_JOBS=${max_jobs}
RUN echo ${TORCH_CUDA_ARCH_LIST} RUN --mount=type=cache,target=/root/.cache/uv bash - <<'BASH'
RUN echo ${MAX_JOBS} export TORCH_CUDA_ARCH_LIST='7.5 8.0+PTX 9.0a'
RUN pip freeze | grep -E 'ninja' git clone https://github.com/facebookresearch/xformers.git
# Build xformers with cuda and torch nightly/wheel pushd xformers
# following official xformers guidance: https://github.com/facebookresearch/xformers#build git checkout v0.0.32.post2
# sha for https://github.com/facebookresearch/xformers/tree/v0.0.32.post2 git submodule update --init --recursive
ARG XFORMERS_COMMIT=5d4b92a5e5a9c6c6d4878283f47d82e17995b468 python3 setup.py bdist_wheel --dist-dir=../xformers-dist --verbose
ENV CCACHE_DIR=/root/.cache/ccache popd
RUN --mount=type=cache,target=/root/.cache/ccache \ rm -rf xformers
--mount=type=cache,target=/root/.cache/uv \ BASH
echo 'git clone xformers...' \
&& git clone https://github.com/facebookresearch/xformers.git --recursive \
&& cd xformers \
&& git checkout ${XFORMERS_COMMIT} \
&& git submodule update --init --recursive \
&& echo 'finish git clone xformers...' \
&& rm -rf build \
&& python3 setup.py bdist_wheel --dist-dir=../xformers-dist --verbose \
&& cd .. \
&& rm -rf xformers
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system xformers-dist/*.whl --verbose uv pip install --system xformers-dist/*.whl
# Build can take a long time, and the torch nightly version fetched from url can be different in next docker stage.
# track the nightly torch version used in the build, when we set up runtime environment we can make sure the version is the same
RUN uv pip freeze | grep -i '^torch\|^torchvision\|^torchaudio' > torch_build_versions.txt RUN uv pip freeze | grep -i '^torch\|^torchvision\|^torchaudio' > torch_build_versions.txt
RUN cat torch_build_versions.txt RUN cat torch_build_versions.txt
RUN pip freeze | grep -E 'torch|xformers|torchvision|torchaudio' RUN pip freeze | grep -E 'torch|xformers|torchvision|torchaudio'
#################### BASE BUILD IMAGE #################### #################### BASE BUILD IMAGE ####################
#################### WHEEL BUILD IMAGE #################### #################### WHEEL BUILD IMAGE ####################
# Image used to build vllm wheel
FROM base AS build FROM base AS build
ARG TARGETPLATFORM ARG TARGETPLATFORM
COPY . . COPY . .
RUN python3 use_existing_torch.py RUN python3 use_existing_torch.py
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
@ -197,20 +155,17 @@ ARG GIT_REPO_CHECK=0
RUN --mount=type=bind,source=.git,target=.git \ RUN --mount=type=bind,source=.git,target=.git \
if [ "$GIT_REPO_CHECK" != "0" ]; then bash tools/check_repo.sh ; fi if [ "$GIT_REPO_CHECK" != "0" ]; then bash tools/check_repo.sh ; fi
# Max jobs used by Ninja to build extensions
ARG max_jobs=16 ARG max_jobs=16
ENV MAX_JOBS=${max_jobs} ENV MAX_JOBS=${max_jobs}
ARG nvcc_threads=4 ARG nvcc_threads=8
ENV NVCC_THREADS=$nvcc_threads ENV NVCC_THREADS=$nvcc_threads
ARG torch_cuda_arch_list='8.0 8.6 8.9 9.0'
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
ARG USE_SCCACHE ARG USE_SCCACHE
ARG SCCACHE_BUCKET_NAME=vllm-build-sccache ARG SCCACHE_BUCKET_NAME=vllm-build-sccache
ARG SCCACHE_REGION_NAME=us-west-2 ARG SCCACHE_REGION_NAME=us-west-2
ARG SCCACHE_S3_NO_CREDENTIALS=0 ARG SCCACHE_S3_NO_CREDENTIALS=0
# if USE_SCCACHE is set, use sccache to speed up compilation # Use sccache to speed up compilation
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
--mount=type=bind,source=.git,target=.git \ --mount=type=bind,source=.git,target=.git \
if [ "$USE_SCCACHE" = "1" ]; then \ if [ "$USE_SCCACHE" = "1" ]; then \
@ -235,6 +190,9 @@ RUN --mount=type=cache,target=/root/.cache/uv \
&& sccache --show-stats; \ && sccache --show-stats; \
fi fi
ARG torch_cuda_arch_list='8.0 8.6 8.9 9.0'
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
ARG vllm_target_device="cuda" ARG vllm_target_device="cuda"
ENV VLLM_TARGET_DEVICE=${vllm_target_device} ENV VLLM_TARGET_DEVICE=${vllm_target_device}
ENV CCACHE_DIR=/root/.cache/ccache ENV CCACHE_DIR=/root/.cache/ccache
@ -248,17 +206,10 @@ RUN --mount=type=cache,target=/root/.cache/ccache \
export VLLM_DOCKER_BUILD_CONTEXT=1 && \ export VLLM_DOCKER_BUILD_CONTEXT=1 && \
python3 setup.py bdist_wheel --dist-dir=vllm-dist --py-limited-api=cp38; \ python3 setup.py bdist_wheel --dist-dir=vllm-dist --py-limited-api=cp38; \
fi fi
RUN echo "[INFO] Listing current directory:" && \
ls -al && \
echo "[INFO] Showing torch_build_versions.txt content:" && \
cat torch_build_versions.txt
#################### WHEEL BUILD IMAGE #################### #################### WHEEL BUILD IMAGE ####################
################### VLLM INSTALLED IMAGE #################### ################### VLLM INSTALLED IMAGE ####################
# Setup clean environment for vLLM for test and api server using ubuntu22.04 with AOT flashinfer
FROM ${FINAL_BASE_IMAGE} AS vllm-base FROM ${FINAL_BASE_IMAGE} AS vllm-base
USER root USER root
@ -266,7 +217,7 @@ ARG CUDA_VERSION
ARG PYTHON_VERSION ARG PYTHON_VERSION
ARG GET_PIP_URL ARG GET_PIP_URL
# TODO (huydhn): Only work with PyTorch manylinux builder # Only work with PyTorch manylinux builder
ENV PATH="/opt/python/cp312-cp312/bin:${PATH}" ENV PATH="/opt/python/cp312-cp312/bin:${PATH}"
# prepare for environment starts # prepare for environment starts
@ -275,20 +226,19 @@ WORKDIR /workspace
# Install Python and other dependencies # Install Python and other dependencies
RUN if command -v apt-get >/dev/null; then \ RUN if command -v apt-get >/dev/null; then \
apt-get update -y \ apt-get update -y \
&& apt-get install -y ccache software-properties-common git curl wget sudo vim \ && apt-get install -y ccache software-properties-common git sudo vim python3-pip; \
&& add-apt-repository -y ppa:deadsnakes/ppa \
&& apt-get update -y \
&& apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \
&& update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
&& update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \
&& ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \
&& curl -sS ${GET_PIP_URL} | python${PYTHON_VERSION}; \
else \ else \
dnf install -y git curl wget sudo; \ dnf install -y git wget sudo; \
fi \ fi \
&& curl -LsSf https://astral.sh/uv/install.sh | sh \
&& $HOME/.local/bin/uv venv /opt/venv --python ${PYTHON_VERSION} \
&& rm -f /usr/bin/python3 /usr/bin/python3-config /usr/bin/pip \
&& ln -s /opt/venv/bin/python3 /usr/bin/python3 \
&& ln -s /opt/venv/bin/python3-config /usr/bin/python3-config \
&& ln -s /opt/venv/bin/pip /usr/bin/pip \
&& python3 --version && python3 -m pip --version && python3 --version && python3 -m pip --version
# Get the torch versions, and whls used in previous stagtes for consistency # Get the torch versions, and whls used in previous stage
COPY --from=base /workspace/torch_build_versions.txt ./torch_build_versions.txt COPY --from=base /workspace/torch_build_versions.txt ./torch_build_versions.txt
COPY --from=base /workspace/xformers-dist /wheels/xformers COPY --from=base /workspace/xformers-dist /wheels/xformers
COPY --from=build /workspace/vllm-dist /wheels/vllm COPY --from=build /workspace/vllm-dist /wheels/vllm
@ -297,33 +247,29 @@ RUN echo "[INFO] Listing current directory before torch install step:" && \
echo "[INFO] Showing torch_build_versions.txt content:" && \ echo "[INFO] Showing torch_build_versions.txt content:" && \
cat torch_build_versions.txt cat torch_build_versions.txt
# Install build and runtime dependencies, this is needed for flashinfer install
COPY requirements/build.txt requirements/build.txt
COPY use_existing_torch.py use_existing_torch.py
RUN python3 use_existing_torch.py
RUN cat requirements/build.txt
# Install uv for faster pip installs if not existed # Install uv for faster pip installs if not existed
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
if ! python3 -m uv --version > /dev/null 2>&1; then \ python3 -m pip install uv==0.8.4
python3 -m pip install uv==0.8.4; \
fi
ENV UV_HTTP_TIMEOUT=500 ENV UV_HTTP_TIMEOUT=500
ENV UV_INDEX_STRATEGY="unsafe-best-match" ENV UV_INDEX_STRATEGY="unsafe-best-match"
# Use copy mode to avoid hardlink failures with Docker cache mounts # Use copy mode to avoid hardlink failures with Docker cache mounts
ENV UV_LINK_MODE=copy ENV UV_LINK_MODE=copy
# Install build and runtime dependencies, this is needed for flashinfer install
COPY requirements/build.txt requirements/build.txt
COPY use_existing_torch.py use_existing_torch.py
RUN python3 use_existing_torch.py
RUN cat requirements/build.txt
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system -r requirements/build.txt uv pip install --system -r requirements/build.txt
# Default mount file as placeholder, this just avoid the mount error # Default mount file as placeholder, this just avoid the mount error
ARG TORCH_WHEELS_PATH="./requirements" ARG TORCH_WHEELS_PATH="./requirements"
# Install torch, torchaudio and torchvision # Install torch, torchaudio and torchvision. If TORCH_WHEELS_PATH is default
# if TORCH_WHEELS_PATH is default "./requirements", it will pull the nightly versions using pip using torch_build_versions.txt # to ./requirements, it will pull the nightly versions using pip. Otherwise,
# otherwise, it will use the whls from TORCH_WHEELS_PATH from the host machine # it will use the local wheels from TORCH_WHEELS_PATH
RUN --mount=type=bind,source=${TORCH_WHEELS_PATH},target=/dist \ RUN --mount=type=bind,source=${TORCH_WHEELS_PATH},target=/dist \
--mount=type=cache,target=/root/.cache/uv \ --mount=type=cache,target=/root/.cache/uv \
if [ -n "$TORCH_WHEELS_PATH" ] && [ "$TORCH_WHEELS_PATH" != "./requirements" ] && [ -d "/dist" ] && ls /dist/torch*.whl >/dev/null 2>&1; then \ if [ -n "$TORCH_WHEELS_PATH" ] && [ "$TORCH_WHEELS_PATH" != "./requirements" ] && [ -d "/dist" ] && ls /dist/torch*.whl >/dev/null 2>&1; then \
@ -344,18 +290,14 @@ RUN --mount=type=cache,target=/root/.cache/uv \
# Install xformers wheel from previous stage # Install xformers wheel from previous stage
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system /wheels/xformers/*.whl --verbose uv pip install --system /wheels/xformers/*.whl --verbose
# Build flashinfer from source.
# Build FlashInfer from source
ARG torch_cuda_arch_list='8.0;8.9;9.0a;10.0a;12.0' ARG torch_cuda_arch_list='8.0;8.9;9.0a;10.0a;12.0'
# install package for build flashinfer
# see issue: https://github.com/flashinfer-ai/flashinfer/issues/738
RUN pip freeze | grep -E 'setuptools|packaging|build'
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list} ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
# Build flashinfer for torch nightly from source around 10 mins
ARG FLASHINFER_GIT_REPO="https://github.com/flashinfer-ai/flashinfer.git" ARG FLASHINFER_GIT_REPO="https://github.com/flashinfer-ai/flashinfer.git"
# Keep this in sync with https://github.com/vllm-project/vllm/blob/main/requirements/cuda.txt
ARG FLASHINFER_GIT_REF="v0.2.14.post1" ARG FLASHINFER_GIT_REF="v0.2.14.post1"
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
git clone --depth 1 --recursive --shallow-submodules \ git clone --depth 1 --recursive --shallow-submodules \
--branch ${FLASHINFER_GIT_REF} \ --branch ${FLASHINFER_GIT_REF} \
@ -367,7 +309,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
&& cd .. \ && cd .. \
&& rm -rf flashinfer && rm -rf flashinfer
# install flashinfer python # Install FlashInfer
RUN --mount=type=cache,target=/root/.cache/uv \ RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system wheels/flashinfer/*.whl --verbose uv pip install --system wheels/flashinfer/*.whl --verbose
@ -377,49 +319,6 @@ RUN uv pip freeze | grep -i '^torch\|^torchvision\|^torchaudio\|^xformers\|^vllm
################### VLLM INSTALLED IMAGE #################### ################### VLLM INSTALLED IMAGE ####################
#################### UNITTEST IMAGE #############################
FROM vllm-base as test
ENV UV_HTTP_TIMEOUT=500
ENV UV_INDEX_STRATEGY="unsafe-best-match"
# Use copy mode to avoid hardlink failures with Docker cache mounts
ENV UV_LINK_MODE=copy
COPY tests/ tests/
COPY examples examples
COPY benchmarks benchmarks
COPY ./vllm/collect_env.py .
COPY requirements/common.txt requirements/common.txt
COPY use_existing_torch.py use_existing_torch.py
COPY pyproject.toml pyproject.toml
# Install build and runtime dependencies without stable torch version
COPY requirements/nightly_torch_test.txt requirements/nightly_torch_test.txt
RUN python3 use_existing_torch.py
# install packages
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system -r requirements/common.txt
# enable fast downloads from hf (for testing)
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system hf_transfer
ENV HF_HUB_ENABLE_HF_TRANSFER 1
# install development dependencies (for testing)
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system -e tests/vllm_test_utils
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install --system -r requirements/nightly_torch_test.txt
# Logging to confirm the torch versions
RUN pip freeze | grep -E 'torch|xformers|vllm|flashinfer'
# Logging to confirm all the packages are installed
RUN pip freeze
#################### UNITTEST IMAGE #############################
#################### EXPORT STAGE #################### #################### EXPORT STAGE ####################
FROM scratch as export-wheels FROM scratch as export-wheels

4
.github/workflows/vllm.yml vendored
View File

@ -46,7 +46,7 @@ jobs:
runner: linux.24xlarge.memory runner: linux.24xlarge.memory
test-matrix: | test-matrix: |
{ include: [ { include: [
{ config: "vllm_basic_correctness_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_basic_correctness_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
{ config: "vllm_basic_models_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_basic_models_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
{ config: "vllm_entrypoints_test", shard: 1, num_shards: 1,runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_entrypoints_test", shard: 1, num_shards: 1,runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
{ config: "vllm_regression_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_regression_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
@ -54,7 +54,7 @@ jobs:
{ config: "vllm_pytorch_compilation_unit_tests", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_pytorch_compilation_unit_tests", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
{ config: "vllm_lora_28_failure_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_lora_28_failure_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
{ config: "vllm_multi_model_test_28_failure_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu"}, { config: "vllm_multi_model_test_28_failure_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu"},
{ config: "vllm_languagde_model_test_extended_generation_28_failure_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu"}, { config: "vllm_language_model_test_extended_generation_28_failure_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu"},
{ config: "vllm_distributed_test_2_gpu_28_failure_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_distributed_test_2_gpu_28_failure_test", shard: 1, num_shards: 1, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
{ config: "vllm_lora_test", shard: 0, num_shards: 4, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_lora_test", shard: 0, num_shards: 4, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },
{ config: "vllm_lora_test", shard: 1, num_shards: 4, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" }, { config: "vllm_lora_test", shard: 1, num_shards: 4, runner: "linux.g6.4xlarge.experimental.nvidia.gpu" },