Skip to content

Commit ab7f130

Browse files
npanpaliyaShafi-Hussain
npanpaliya
and
Shafi-Hussain
authored
UBI9 based Dockerfile for IBM ppc64le (#342)
UBI9 based Dockerfile for ppc64le which builds vllm's dependencies from source. Co-authored-by: Md. Shafi Hussain <[email protected]>
1 parent 27f0118 commit ab7f130

File tree

4 files changed

+243
-1
lines changed

4 files changed

+243
-1
lines changed

‎Dockerfile.ppc64le.ubi

Lines changed: 239 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,239 @@
1+
ARG BASE_UBI_IMAGE_TAG=9.5-1741599792
2+
###############################################################
3+
# base stage with basic dependencies
4+
###############################################################
5+
FROM registry.access.redhat.com/ubi9/ubi-minimal:${BASE_UBI_IMAGE_TAG} AS base-builder
6+
ARG PYTHON_VERSION=3.12
7+
ARG OPENBLAS_VERSION=0.3.29
8+
# Set Environment Variables for venv, cargo & openblas
9+
ENV VIRTUAL_ENV=/opt/vllm
10+
ENV PATH=${VIRTUAL_ENV}/bin:/root/.cargo/bin:$PATH
11+
ENV PKG_CONFIG_PATH=/usr/local/lib/pkgconfig/
12+
ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib64:/usr/local/lib:/usr/lib64:/usr/lib
13+
# install gcc-13, python, rust, openblas
14+
# Note: A symlink for libatomic.so is created for gcc-13 (linker fails to find to find libatomic otherwise - reqd. for sentencepiece)
15+
# Note: A dummy file 'control' is created in /tmp/ to artifically create dependencies between stages when building stages in parallel
16+
# when `--jobs=<N>` is passed with podman build command
17+
RUN microdnf install -y dnf \
18+
&& dnf install -y https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-gpg-keys-9.0-24.el9.noarch.rpm \
19+
https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os/Packages/centos-stream-repos-9.0-24.el9.noarch.rpm \
20+
https://dl.fedoraproject.org/pub/epel/epel-release-latest-9.noarch.rpm \
21+
&& dnf config-manager --add-repo https://mirror.stream.centos.org/9-stream/BaseOS/`arch`/os \
22+
&& dnf config-manager --add-repo https://mirror.stream.centos.org/9-stream/AppStream/`arch`/os \
23+
&& dnf config-manager --set-enabled crb \
24+
&& dnf install -y \
25+
git tar gcc-toolset-13 numactl-devel lapack-devel \
26+
pkgconfig openssl-devel xsimd zeromq-devel kmod \
27+
libtiff-devel libjpeg-devel openjpeg2-devel zlib-devel \
28+
freetype-devel lcms2-devel libwebp-devel tcl-devel tk-devel \
29+
harfbuzz-devel fribidi-devel libraqm-devel libimagequant-devel libxcb-devel \
30+
python${PYTHON_VERSION}-devel python${PYTHON_VERSION}-pip \
31+
&& dnf clean all \
32+
&& ln -sf /usr/lib64/libatomic.so.1 /usr/lib64/libatomic.so \
33+
&& python${PYTHON_VERSION} -m venv ${VIRTUAL_ENV} \
34+
&& python -m pip install -U pip uv \
35+
&& uv pip install wheel build setuptools setuptools_scm setuptools_rust meson-python cmake ninja cython scikit_build_core scikit_build \
36+
&& curl -sL https://ftp2.osuosl.org/pub/ppc64el/openblas/latest/Openblas_${OPENBLAS_VERSION}_ppc64le.tar.gz | tar xvf - -C /usr/local \
37+
&& curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y \
38+
&& cd /tmp && touch control
39+
###############################################################
40+
# Stage to build torch family
41+
###############################################################
42+
FROM base-builder AS torch-builder
43+
# buld cache without torch dependent packages
44+
# sentencepiece has linker issues otherwise
45+
RUN --mount=type=cache,target=/root/.cache/uv \
46+
--mount=type=bind,src=requirements-common.txt,dst=/requirements-common.txt,ro \
47+
source /opt/rh/gcc-toolset-13/enable && \
48+
sed -e 's/.*opencv.*//g; s/.*outlines.*//g; s/.*tensors.*//g' requirements-common.txt | uv pip install -r /dev/stdin && \
49+
uv pip list
50+
ARG MAX_JOBS
51+
ARG TORCH_VERSION=2.5.1
52+
ARG _GLIBCXX_USE_CXX11_ABI=1
53+
RUN --mount=type=cache,target=/root/.cache/uv \
54+
source /opt/rh/gcc-toolset-13/enable && \
55+
git clone --recursive https://github.com/pytorch/pytorch.git -b v${TORCH_VERSION} && \
56+
cd pytorch && \
57+
uv pip install -r requirements.txt && \
58+
python setup.py develop && \
59+
rm -f dist/torch*+git*whl && \
60+
MAX_JOBS=${MAX_JOBS:-$(nproc)} \
61+
PYTORCH_BUILD_VERSION=${TORCH_VERSION} PYTORCH_BUILD_NUMBER=1 uv build --wheel --out-dir /torchwheels/
62+
ARG TORCHVISION_VERSION=0.20.1
63+
ARG TORCHVISION_USE_NVJPEG=0
64+
ARG TORCHVISION_USE_FFMPEG=0
65+
RUN --mount=type=cache,target=/root/.cache/uv \
66+
source /opt/rh/gcc-toolset-13/enable && \
67+
git clone --recursive https://github.com/pytorch/vision.git -b v${TORCHVISION_VERSION} && \
68+
cd vision && \
69+
MAX_JOBS=${MAX_JOBS:-$(nproc)} \
70+
BUILD_VERSION=${TORCHVISION_VERSION} \
71+
uv build --wheel --out-dir /torchwheels/ --no-build-isolation
72+
ARG TORCHAUDIO_VERSION=2.5.1
73+
ARG BUILD_SOX=1
74+
ARG BUILD_KALDI=1
75+
ARG BUILD_RNNT=1
76+
ARG USE_FFMPEG=0
77+
ARG USE_ROCM=0
78+
ARG USE_CUDA=0
79+
ARG TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_FFMPEG=1
80+
RUN --mount=type=cache,target=/root/.cache/uv \
81+
source /opt/rh/gcc-toolset-13/enable && \
82+
git clone --recursive https://github.com/pytorch/audio.git -b v${TORCHAUDIO_VERSION} && \
83+
cd audio && \
84+
MAX_JOBS=${MAX_JOBS:-$(nproc)} \
85+
BUILD_VERSION=${TORCHAUDIO_VERSION} \
86+
uv build --wheel --out-dir /torchwheels/ --no-build-isolation
87+
# build cache with torch dependent packages
88+
RUN --mount=type=cache,target=/root/.cache/uv \
89+
--mount=type=bind,src=requirements-common.txt,dst=/requirements-common.txt,ro \
90+
source /opt/rh/gcc-toolset-13/enable && \
91+
grep "outlines" requirements-common.txt | uv pip install -r /dev/stdin && \
92+
grep "tensors" requirements-common.txt | uv pip install -r /dev/stdin && \
93+
uv pip list
94+
###############################################################
95+
# Stage to build pyarrow
96+
###############################################################
97+
FROM base-builder AS arrow-builder
98+
ARG MAX_JOBS
99+
ARG PYARROW_PARALLEL
100+
ARG PYARROW_VERSION=19.0.1
101+
RUN --mount=type=cache,target=/root/.cache/uv \
102+
source /opt/rh/gcc-toolset-13/enable && \
103+
git clone --recursive https://github.com/apache/arrow.git -b apache-arrow-${PYARROW_VERSION} && \
104+
cd arrow/cpp && \
105+
mkdir build && cd build && \
106+
cmake -DCMAKE_BUILD_TYPE=release \
107+
-DCMAKE_INSTALL_PREFIX=/usr/local \
108+
-DARROW_PYTHON=ON \
109+
-DARROW_BUILD_TESTS=OFF \
110+
-DARROW_JEMALLOC=ON \
111+
-DARROW_BUILD_STATIC="OFF" \
112+
-DARROW_PARQUET=ON \
113+
.. && \
114+
make install -j ${MAX_JOBS:-$(nproc)} && \
115+
cd ../../python/ && \
116+
uv pip install -v -r requirements-wheel-build.txt && \
117+
PYARROW_PARALLEL=${PYARROW_PARALLEL:-$(nproc)} \
118+
python setup.py build_ext \
119+
--build-type=release --bundle-arrow-cpp \
120+
bdist_wheel --dist-dir /arrowwheels/
121+
###############################################################
122+
# Stage to build opencv
123+
###############################################################
124+
FROM base-builder AS cv-builder
125+
ARG MAX_JOBS
126+
ARG OPENCV_VERSION=84
127+
ARG ENABLE_HEADLESS=1
128+
RUN --mount=type=cache,target=/root/.cache/uv \
129+
source /opt/rh/gcc-toolset-13/enable && \
130+
git clone --recursive https://github.com/opencv/opencv-python.git -b ${OPENCV_VERSION} && \
131+
cd opencv-python && \
132+
sed -i 's/"setuptools==59.2.0",/"setuptools<70.0",/g' pyproject.toml && \
133+
python -m build --wheel --installer=uv --outdir /opencvwheels/
134+
# build opencv dependent packages
135+
RUN --mount=type=cache,target=/root/.cache/uv \
136+
--mount=type=bind,src=requirements-common.txt,dst=/requirements-common.txt,ro \
137+
source /opt/rh/gcc-toolset-13/enable && \
138+
uv pip install /opencvwheels/*.whl && \
139+
uv pip list | grep opencv && \
140+
grep "opencv" requirements-common.txt | uv pip install -r /dev/stdin --no-build-isolation && \
141+
uv pip list
142+
###############################################################
143+
# Stage to build vllm - this stage builds and installs
144+
# vllm, tensorizer and vllm-tgis-adapter and builds uv cache
145+
# for transitive dependencies - eg. grpcio
146+
###############################################################
147+
FROM base-builder AS vllmcache-builder
148+
COPY --from=torch-builder /tmp/control /dev/null
149+
COPY --from=arrow-builder /tmp/control /dev/null
150+
COPY --from=cv-builder /tmp/control /dev/null
151+
ARG VLLM_TGIS_ADAPTER_VERSION=0.6.3
152+
ARG VLLM_TARGET_DEVICE=cpu
153+
ARG GRPC_PYTHON_BUILD_SYSTEM_OPENSSL=1
154+
# this step installs vllm & tgis adapter and populates uv cache
155+
# with all the transitive dependencies
156+
RUN --mount=type=cache,target=/root/.cache/uv \
157+
--mount=type=bind,from=torch-builder,source=/torchwheels/,target=/torchwheels/,ro \
158+
--mount=type=bind,from=arrow-builder,source=/arrowwheels/,target=/arrowwheels/,ro \
159+
--mount=type=bind,from=cv-builder,source=/opencvwheels/,target=/opencvwheels/,ro \
160+
--mount=type=bind,src=.,dst=/src/,rw \
161+
source /opt/rh/gcc-toolset-13/enable && \
162+
uv pip install /opencvwheels/*.whl /arrowwheels/*.whl /torchwheels/*.whl && \
163+
sed -i -e 's/.*torch.*//g' /src/pyproject.toml /src/requirements-*.txt && \
164+
uv pip install pandas && \
165+
uv pip install -r /src/requirements-common.txt -r /src/requirements-cpu.txt -r /src/requirements-build.txt --no-build-isolation && \
166+
cd /src/ && \
167+
uv build --wheel --out-dir /vllmwheel/ --no-build-isolation && \
168+
uv pip install "$(echo /vllmwheel/*.whl)[tensorizer]" vllm-tgis-adapter==${VLLM_TGIS_ADAPTER_VERSION}
169+
###############################################################
170+
# FINAL VLLM IMAGE STAGE #
171+
###############################################################
172+
FROM registry.access.redhat.com/ubi9/ubi-minimal:${BASE_UBI_IMAGE_TAG} AS vllm-openai
173+
ARG PYTHON_VERSION=3.12
174+
ARG OPENBLAS_VERSION=0.3.29
175+
# Set Environment Variables for venv & openblas
176+
ENV VIRTUAL_ENV=/opt/vllm
177+
ENV PATH=${VIRTUAL_ENV}/bin:$PATH
178+
ENV PKG_CONFIG_PATH=/usr/local/lib/pkgconfig/
179+
ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib64:/usr/local/lib:/usr/lib64:/usr/lib
180+
# install gcc-13, python, openblas
181+
RUN rpm -ivh https://dl.fedoraproject.org/pub/epel/epel-release-latest-9.noarch.rpm && \
182+
microdnf install --nodocs -y \
183+
tar findutils openssl numactl lapack \
184+
pkgconfig xsimd g++ libsndfile \
185+
libtiff libjpeg openjpeg2 zlib zeromq \
186+
freetype lcms2 libwebp tcl tk utf8proc \
187+
harfbuzz fribidi libraqm libimagequant libxcb \
188+
python${PYTHON_VERSION}-devel python${PYTHON_VERSION}-pip \
189+
&& microdnf clean all \
190+
&& python${PYTHON_VERSION} -m venv ${VIRTUAL_ENV} \
191+
&& python -m pip install -U pip uv --no-cache \
192+
&& curl -sL https://ftp2.osuosl.org/pub/ppc64el/openblas/latest/Openblas_${OPENBLAS_VERSION}_ppc64le.tar.gz | tar xvf - -C /usr/local
193+
# create artificial dependencies between stages for independent stages to build in parallel
194+
COPY --from=torch-builder /tmp/control /dev/null
195+
COPY --from=arrow-builder /tmp/control /dev/null
196+
COPY --from=cv-builder /tmp/control /dev/null
197+
COPY --from=vllmcache-builder /tmp/control /dev/null
198+
# consume previously built wheels
199+
RUN --mount=type=cache,target=/root/.cache/uv \
200+
--mount=type=bind,from=torch-builder,source=/torchwheels/,target=/torchwheels/,ro \
201+
--mount=type=bind,from=arrow-builder,source=/arrowwheels/,target=/arrowwheels/,ro \
202+
--mount=type=bind,from=cv-builder,source=/opencvwheels/,target=/opencvwheels/,ro \
203+
--mount=type=bind,from=vllmcache-builder,source=/vllmwheel/,target=/vllmwheel/,ro \
204+
HOME=/root uv pip install /opencvwheels/*.whl /arrowwheels/*.whl /torchwheels/*.whl /vllmwheel/*.whl
205+
WORKDIR /home/vllm
206+
# setup non-root user for OpenShift
207+
RUN umask 002 && \
208+
useradd --uid 2000 --gid 0 vllm && \
209+
mkdir -p /home/vllm && \
210+
chmod g+rwx /home/vllm
211+
ENV HF_HUB_OFFLINE=0 \
212+
HOME=/home/vllm \
213+
# Allow requested max length to exceed what is extracted from the
214+
# config.json
215+
# see: https://github.com/vllm-project/vllm/pull/7080
216+
VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
217+
VLLM_USAGE_SOURCE=production-docker-image \
218+
VLLM_WORKER_MULTIPROC_METHOD=fork \
219+
VLLM_NO_USAGE_STATS=1 \
220+
OUTLINES_CACHE_DIR=/tmp/outlines
221+
COPY LICENSE /licenses/vllm.md
222+
COPY examples/*.jinja /app/data/template/
223+
USER 2000
224+
ENTRYPOINT ["python", "-m", "vllm.entrypoints.openai.api_server"]
225+
FROM vllm-openai as vllm-grpc-adapter
226+
USER root
227+
ARG VLLM_TGIS_ADAPTER_VERSION=0.6.3
228+
RUN --mount=type=cache,target=/root/.cache/uv \
229+
--mount=type=bind,from=vllmcache-builder,source=/vllmwheel/,target=/vllmwheel/,ro \
230+
HOME=/root uv pip install "$(echo /vllmwheel/*.whl)[tensorizer]" vllm-tgis-adapter==${VLLM_TGIS_ADAPTER_VERSION}
231+
ENV GRPC_PORT=8033 \
232+
PORT=8000 \
233+
# As an optimization, vLLM disables logprobs when using spec decoding by
234+
# default, but this would be unexpected to users of a hosted model that
235+
# happens to have spec decoding
236+
# see: https://github.com/vllm-project/vllm/pull/6485
237+
DISABLE_LOGPROBS_DURING_SPEC_DECODING=false
238+
USER 2000
239+
ENTRYPOINT ["python", "-m", "vllm_tgis_adapter", "--uvicorn-log-level=warning"]

‎requirements-common.txt

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,6 @@
11
psutil
22
sentencepiece # Required for LLaMA tokenizer.
33
numpy < 2.0.0
4-
numba == 0.60.0 # v0.61 doesn't support Python 3.9. Required for N-gram speculative decoding.
54
requests >= 2.26.0
65
tqdm
76
blake3

‎requirements-cuda.txt

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,8 @@
11
# Common dependencies
22
-r requirements-common.txt
33

4+
numba == 0.60.0 # v0.61 doesn't support Python 3.9. Required for N-gram speculative decoding.
5+
46
# Dependencies for NVIDIA GPUs
57
ray[adag] == 2.40.0 # Required for pipeline parallelism in V1.
68
torch == 2.5.1

‎requirements-rocm.txt

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,8 @@
11
# Common dependencies
22
-r requirements-common.txt
33

4+
numba == 0.60.0 # v0.61 doesn't support Python 3.9. Required for N-gram speculative decoding.
5+
46
# Dependencies for AMD GPUs
57
awscli
68
boto3

0 commit comments

Comments
 (0)