Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[CI/Build] Dockerfile build for ARM64 / GH200 #10499

Closed
wants to merge 4 commits into from
Closed

[CI/Build] Dockerfile build for ARM64 / GH200 #10499

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d2cbe42
Update Docker for aarch64 builds
drikster80 Nov 20, 2024
29ed358
Update docs for arm64 docker builds & GH200 example
drikster80 Nov 20, 2024
f2635a4
Fix FLASHINFER not installing by default on x86_64
drikster80 Nov 20, 2024
ccef455
Removed build of flash-attn. Added libnccl2
drikster80 Dec 4, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
78 changes: 72 additions & 6 deletions Dockerfile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,13 +11,14 @@ ARG CUDA_VERSION=12.4.1
FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu20.04 AS base
ARG CUDA_VERSION=12.4.1
ARG PYTHON_VERSION=3.12
ARG TARGETPLATFORM
ENV DEBIAN_FRONTEND=noninteractive

# Install Python and other dependencies
RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
&& echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections \
&& apt-get update -y \
&& apt-get install -y ccache software-properties-common git curl sudo \
&& apt-get install -y ccache software-properties-common git curl sudo kmod \
&& add-apt-repository ppa:deadsnakes/ppa \
&& apt-get update -y \
&& apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \
Expand Down Expand Up @@ -46,9 +47,14 @@ WORKDIR /workspace
# install build and runtime dependencies
COPY requirements-common.txt requirements-common.txt
COPY requirements-cuda.txt requirements-cuda.txt
COPY requirements-cuda-arm64.txt requirements-cuda-arm64.txt
RUN --mount=type=cache,target=/root/.cache/pip \
python3 -m pip install -r requirements-cuda.txt

RUN --mount=type=cache,target=/root/.cache/pip \
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
python3 -m pip install -r requirements-cuda-arm64.txt; \
fi

# cuda arch list used by torch
# can be useful for both `dev` and `test`
Expand All @@ -63,13 +69,19 @@ ENV VLLM_FA_CMAKE_GPU_ARCHES=${vllm_fa_cmake_gpu_arches}

#################### WHEEL BUILD IMAGE ####################
FROM base AS build
ARG TARGETPLATFORM

# install build dependencies
COPY requirements-build.txt requirements-build.txt

RUN --mount=type=cache,target=/root/.cache/pip \
python3 -m pip install -r requirements-build.txt

RUN --mount=type=cache,target=/root/.cache/pip \
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
python3 -m pip install -r requirements-cuda-arm64.txt; \
fi

COPY . .
ARG GIT_REPO_CHECK=0
RUN --mount=type=bind,source=.git,target=.git \
Expand Down Expand Up @@ -113,6 +125,47 @@ RUN --mount=type=cache,target=/root/.cache/ccache \
python3 setup.py bdist_wheel --dist-dir=dist --py-limited-api=cp38; \
fi

RUN --mount=type=cache,target=/root/.cache/pip \
--mount=type=bind,source=.git,target=.git \
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
apt-get update && apt-get install zlib1g-dev && \
python3 -m pip install packaging pybind11 && \
git clone https://github.com/openai/triton && \
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can we directly use pytorch nightly as base image so that we don't need to build triton, etc?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm confused. Triton doesn't provide aarch64 whl files, so we'll always need to compile it if we want to use the latest version: https://pypi.org/project/triton/#files

It probably is a good idea to pin to the latest release tag of triton, instead of the main though. I'll update that.

My goal on this was to keep it as close as possible to the x86_64 implementation of VLLM, so I didn't want to use the nvidia pytorch container. That's what I was doing in the previous repo. Although it worked, it doubled the size of the final image (9.74GB vs 4.89GB).

cd triton/python && \
git submodule update --init --recursive && \
pip --verbose wheel --use-pep517 --no-deps -w /workspace/dist --no-build-isolation --no-cache-dir . ; \
fi

RUN --mount=type=cache,target=/root/.cache/pip \
--mount=type=bind,source=.git,target=.git \
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
CAUSAL_CONV1D_FORCE_BUILD=TRUE CAUSAL_CONV1D_SKIP_CUDA_BUILD=FALSE pip --verbose wheel --use-pep517 --no-deps -w /workspace/dist --no-build-isolation --no-cache-dir git+https://github.com/Dao-AILab/causal-conv1d.git ; \
fi

RUN --mount=type=cache,target=/root/.cache/pip \
--mount=type=bind,source=.git,target=.git \
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
MAMBA_FORCE_BUILD=TRUE pip --verbose wheel --use-pep517 --no-deps -w /workspace/dist --no-build-isolation --no-cache-dir git+https://github.com/state-spaces/mamba.git ; \
fi

RUN --mount=type=cache,target=/root/.cache/pip \
--mount=type=bind,source=.git,target=.git \
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
apt-get update && apt-get install -y cuda-toolkit-12-4 && \
git clone -b v0.1.6 https://github.com/flashinfer-ai/flashinfer.git --recursive && \
cd flashinfer/python && \
pip --verbose wheel --use-pep517 --no-deps -w /workspace/dist --no-build-isolation --no-cache-dir . ; \
fi

RUN --mount=type=cache,target=/root/.cache/pip \
--mount=type=bind,source=.git,target=.git \
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
git clone -b 0.44.1 https://github.com/bitsandbytes-foundation/bitsandbytes.git && \
cd bitsandbytes && \
pip --verbose wheel --use-pep517 --no-deps -w /workspace/dist --no-build-isolation --no-cache-dir . ; \
fi


# Check the size of the wheel if RUN_WHEEL_CHECK is true
COPY .buildkite/check-wheel-size.py check-wheel-size.py
# Default max size of the wheel is 250MB
Expand All @@ -124,6 +177,7 @@ RUN if [ "$RUN_WHEEL_CHECK" = "true" ]; then \
else \
echo "Skipping wheel size check."; \
fi

#################### EXTENSION Build IMAGE ####################

#################### DEV IMAGE ####################
Expand All @@ -143,6 +197,9 @@ ARG CUDA_VERSION=12.4.1
ARG PYTHON_VERSION=3.12
WORKDIR /vllm-workspace
ENV DEBIAN_FRONTEND=noninteractive
ARG TARGETPLATFORM

COPY requirements-cuda-arm64.txt requirements-cuda-arm64.txt

RUN PYTHON_VERSION_STR=$(echo ${PYTHON_VERSION} | sed 's/\.//g') && \
echo "export PYTHON_VERSION_STR=${PYTHON_VERSION_STR}" >> /etc/environment
Expand All @@ -151,7 +208,7 @@ RUN PYTHON_VERSION_STR=$(echo ${PYTHON_VERSION} | sed 's/\.//g') && \
RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
&& echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections \
&& apt-get update -y \
&& apt-get install -y ccache software-properties-common git curl sudo vim python3-pip \
&& apt-get install -y ccache software-properties-common git curl sudo vim python3-pip libnccl2\
&& apt-get install -y ffmpeg libsm6 libxext6 libgl1 \
&& add-apt-repository ppa:deadsnakes/ppa \
&& apt-get update -y \
Expand All @@ -168,14 +225,23 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
# or future versions of triton.
RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/

# install vllm wheel first, so that torch etc will be installed
# Install vllm wheel first, so that torch etc will be installed.
# On Arm64 platforms, all newly compiled wheels will also be installed (flashinfer, triton, mamba, etc.)
RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist \
--mount=type=cache,target=/root/.cache/pip \
python3 -m pip install dist/*.whl --verbose

RUN --mount=type=cache,target=/root/.cache/pip \
. /etc/environment && \
python3 -m pip install https://github.com/flashinfer-ai/flashinfer/releases/download/v0.1.6/flashinfer-0.1.6+cu121torch2.4-cp${PYTHON_VERSION_STR}-cp${PYTHON_VERSION_STR}-linux_x86_64.whl
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
pip uninstall -y torch && \
python3 -m pip install -r requirements-cuda-arm64.txt; \
fi

RUN --mount=type=cache,target=/root/.cache/pip \
. /etc/environment && \
if [ "$TARGETPLATFORM" != "linux/arm64" ]; then \
python3 -m pip install https://github.com/flashinfer-ai/flashinfer/releases/download/v0.1.6/flashinfer-0.1.6+cu121torch2.4-cp${PYTHON_VERSION_STR}-cp${PYTHON_VERSION_STR}-linux_x86_64.whl; \
fi
COPY examples examples
#################### vLLM installation IMAGE ####################

Expand Down Expand Up @@ -214,7 +280,7 @@ FROM vllm-base AS vllm-openai

# install additional dependencies for openai api server
RUN --mount=type=cache,target=/root/.cache/pip \
pip install accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.44.0' timm==0.9.10
pip install accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.44.0' 'timm==0.9.10'

ENV VLLM_USAGE_SOURCE production-docker-image

Expand Down
26 changes: 26 additions & 0 deletions docs/source/serving/deploying_with_docker.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,6 +37,32 @@ You can build and run vLLM from source via the provided `Dockerfile <https://git
current GPU type the machine is running on, you can add the argument ``--build-arg torch_cuda_arch_list=""``
for vLLM to find the current GPU type and build for that.

Building for Arm64/aarch64
--------------------------

A docker container can be built for aarch64 systems such as the Nvidia Grace-Hopper. At time of this writing, this requires the use
of PyTorch Nightly and should be considered **experimental**. Using the flag `--platform "linux/arm64"` will attempt to build for arm64.

.. note::

Multiple modules must be compiled, so this process can take a while. Recommend using `--build-arg max_jobs=` & `--build-arg nvcc_threads=`
flags to speed up build process. However, ensure your 'max_jobs' is substantially larger than 'nvcc_threads' to get the most benefits.
Keep an eye on memory usage with parallel jobs as it can be substantial (see example below).

.. code-block:: console

# Example of building on Nvidia GH200 server. (Memory usage: ~180GB, Build time: ~2387s / ~40 min)
$ DOCKER_BUILDKIT=1 sudo docker build . \
--target vllm-openai \
-platform "linux/arm64" \
-t drikster80/vllm-gh200-openai:v0.6.4.post1 \
--build-arg max_jobs=66 \
--build-arg nvcc_threads=2 \
--build-arg torch_cuda_arch_list="9.0+PTX" \
--build-arg vllm_fa_cmake_gpu_arches="90-real"




To run vLLM:

Expand Down
2 changes: 1 addition & 1 deletion requirements-build.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,6 @@ ninja
packaging
setuptools>=61
setuptools-scm>=8
torch==2.5.1
torch==2.5.1; platform_machine != 'aarch64'
wheel
jinja2
3 changes: 3 additions & 0 deletions requirements-cuda-arm64.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
--index-url https://download.pytorch.org/whl/nightly/cu124
torchvision; platform_machine == 'aarch64'
torch; platform_machine == 'aarch64'
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can add xformers for aarch64 to the /vllm-project directory similar to flash-attention for the aarch64 build until the upstream pip package is available

4 changes: 2 additions & 2 deletions requirements-cuda.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@
# Dependencies for NVIDIA GPUs
ray >= 2.9
nvidia-ml-py >= 12.560.30 # for pynvml package
torch == 2.5.1
torch == 2.5.1; platform_machine != 'aarch64'
# These must be updated alongside torch
torchvision == 0.20.1 # Required for phi3v processor. See https://github.com/pytorch/vision?tab=readme-ov-file#installation for corresponding version
torchvision == 0.20.1; platform_machine != 'aarch64' # Required for phi3v processor. See https://github.com/pytorch/vision?tab=readme-ov-file#installation for corresponding version
xformers == 0.0.28.post3; platform_system == 'Linux' and platform_machine == 'x86_64' # Requires PyTorch 2.5.1
Loading