diff --git a/orttraining/tools/ci_test/run_batch_size_test.py b/orttraining/tools/ci_test/run_batch_size_test.py index ea82cc918e..b735f71bb7 100755 --- a/orttraining/tools/ci_test/run_batch_size_test.py +++ b/orttraining/tools/ci_test/run_batch_size_test.py @@ -46,8 +46,8 @@ def main(): ] configs['MI100_32G'] = [ - Config(True, 128, 200, 20, ""), - Config(True, 512, 30, 80, ""), + Config(True, 128, 192, 20, ""), + Config(True, 512, 26, 80, ""), Config(False, 128, 108, 20, ""), Config(False, 512, 16, 80, ""), ] diff --git a/tools/ci_build/github/azure-pipelines/templates/py-packaging-stage.yml b/tools/ci_build/github/azure-pipelines/templates/py-packaging-stage.yml index adabdaa8f7..f0ea943195 100644 --- a/tools/ci_build/github/azure-pipelines/templates/py-packaging-stage.yml +++ b/tools/ci_build/github/azure-pipelines/templates/py-packaging-stage.yml @@ -287,7 +287,7 @@ stages: - ${{ if eq(parameters.enable_linux_rocm_training, true) }}: - job: ROCm_build_environment displayName: 'Construct ROCm wheels environment' - timeoutInMinutes: 180 + timeoutInMinutes: 300 workspace: clean: all pool: AMD-GPU @@ -318,6 +318,19 @@ stages: --build-arg PREPEND_PATH=/opt/rh/devtoolset-10/root/usr/bin: --build-arg LD_LIBRARY_PATH_ARG=/opt/rh/devtoolset-10/root/usr/lib64:/opt/rh/devtoolset-10/root/usr/lib:/opt/rh/devtoolset-10/root/usr/lib64/dyninst:/opt/rh/devtoolset-10/root/usr/lib/dyninst:/usr/local/lib64:/usr/local/lib Repository: onnxruntimetrainingrocmbuild-torch1.10.0-rocm4.3.1 + - template: get-docker-image-steps.yml + parameters: + Dockerfile: tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_rocm5_0_1 + Context: tools/ci_build/github/linux/docker + DockerBuildArgs: >- + --build-arg TORCH_VERSION=1.10.0 + --build-arg INSTALL_DEPS_EXTRA_ARGS=-tmur + --build-arg BUILD_UID=$(id -u) + --network=host --build-arg POLICY=manylinux2014 --build-arg PLATFORM=x86_64 + --build-arg DEVTOOLSET_ROOTPATH=/opt/rh/devtoolset-10/root + --build-arg PREPEND_PATH=/opt/rh/devtoolset-10/root/usr/bin: + --build-arg LD_LIBRARY_PATH_ARG=/opt/rh/devtoolset-10/root/usr/lib64:/opt/rh/devtoolset-10/root/usr/lib:/opt/rh/devtoolset-10/root/usr/lib64/dyninst:/opt/rh/devtoolset-10/root/usr/lib/dyninst:/usr/local/lib64:/usr/local/lib + Repository: onnxruntimetrainingrocmbuild-torch1.10.0-rocm5.0.1 - job: ROCM_training_wheels timeoutInMinutes: 180 @@ -353,6 +366,18 @@ stages: PythonVersion: '3.9' TorchVersion: '1.10.0' RocmVersion: '4.3.1' + Python37 Torch1100 Rocm501: + PythonVersion: '3.7' + TorchVersion: '1.10.0' + RocmVersion: '5.0.1' + Python38 Torch1100 Rocm501: + PythonVersion: '3.8' + TorchVersion: '1.10.0' + RocmVersion: '5.0.1' + Python39 Torch1100 Rocm501: + PythonVersion: '3.9' + TorchVersion: '1.10.0' + RocmVersion: '5.0.1' steps: - checkout: self diff --git a/tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_rocm5_0_1 b/tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_rocm5_0_1 new file mode 100644 index 0000000000..c41397e1e4 --- /dev/null +++ b/tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_rocm5_0_1 @@ -0,0 +1,221 @@ +ARG BASEIMAGE=rocm/pytorch:rocm5.0.1_centos7_py3.7_pytorch_staging +ARG POLICY=manylinux2014 +ARG PLATFORM=x86_64 +ARG DEVTOOLSET_ROOTPATH= +ARG LD_LIBRARY_PATH_ARG= +ARG PREPEND_PATH= + +#Build manylinux2014 docker image begin +FROM $BASEIMAGE AS runtime_base +ARG POLICY +ARG PLATFORM +ARG DEVTOOLSET_ROOTPATH +ARG LD_LIBRARY_PATH_ARG +ARG PREPEND_PATH +LABEL maintainer="The ManyLinux project" + +# TODO: This should not be required if the hipify-perl binary is already a part +# of the base docker image. Remove this line moving forward. +RUN yum install -y hipify-clang + +RUN yum remove -y devtoolset\* git\* && conda remove -y cmake + +# remove protobuf 2.6.1 from rocm/pytorch:rocm5.0.1_centos7_py3.7_pytorch_staging +# it's too old to compile onnx 1.10 +RUN rm -fr /usr/local/bin/protoc \ + /usr/local/libproto* \ + /usr/local/include/google \ + /usr/local/lib/pkgconfig/protobuf* + +ENV AUDITWHEEL_POLICY=${POLICY} AUDITWHEEL_ARCH=${PLATFORM} AUDITWHEEL_PLAT=${POLICY}_${PLATFORM} +ENV LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8 LANGUAGE=en_US.UTF-8 +ENV DEVTOOLSET_ROOTPATH=${DEVTOOLSET_ROOTPATH} +ENV LD_LIBRARY_PATH=${LD_LIBRARY_PATH_ARG} +ENV PATH=${PREPEND_PATH}${PATH} +ENV PKG_CONFIG_PATH=/usr/local/lib/pkgconfig + +# first copy the fixup mirrors script, keep the script around +COPY build_scripts/fixup-mirrors.sh /usr/local/sbin/fixup-mirrors + +# setup entrypoint, this will wrap commands with `linux32` with i686 images +COPY build_scripts/install-entrypoint.sh \ + build_scripts/update-system-packages.sh \ + build_scripts/build_utils.sh \ + /build_scripts/ + +RUN /build_scripts/install-entrypoint.sh && rm -rf /build_scripts +COPY manylinux-entrypoint /usr/local/bin/manylinux-entrypoint +ENTRYPOINT ["manylinux-entrypoint"] + +COPY build_scripts/install-runtime-packages.sh \ + build_scripts/update-system-packages.sh \ + build_scripts/build_utils.sh \ + /build_scripts/ +RUN manylinux-entrypoint /build_scripts/install-runtime-packages.sh && rm -rf /build_scripts/ + +COPY build_scripts/build_utils.sh /build_scripts/ + +COPY build_scripts/install-autoconf.sh /build_scripts/ +RUN export AUTOCONF_ROOT=autoconf-2.71 && \ + export AUTOCONF_HASH=431075ad0bf529ef13cb41e9042c542381103e80015686222b8a9d4abef42a1c && \ + export AUTOCONF_DOWNLOAD_URL=http://ftp.gnu.org/gnu/autoconf && \ + manylinux-entrypoint /build_scripts/install-autoconf.sh + +COPY build_scripts/install-automake.sh /build_scripts/ +RUN export AUTOMAKE_ROOT=automake-1.16.5 && \ + export AUTOMAKE_HASH=07bd24ad08a64bc17250ce09ec56e921d6343903943e99ccf63bbf0705e34605 && \ + export AUTOMAKE_DOWNLOAD_URL=http://ftp.gnu.org/gnu/automake && \ + manylinux-entrypoint /build_scripts/install-automake.sh + +COPY build_scripts/install-libtool.sh /build_scripts/ +RUN export LIBTOOL_ROOT=libtool-2.4.6 && \ + export LIBTOOL_HASH=e3bd4d5d3d025a36c21dd6af7ea818a2afcd4dfc1ea5a17b39d7854bcd0c06e3 && \ + export LIBTOOL_DOWNLOAD_URL=http://ftp.gnu.org/gnu/libtool && \ + manylinux-entrypoint /build_scripts/install-libtool.sh + +COPY build_scripts/install-patchelf.sh /build_scripts/ +RUN export PATCHELF_VERSION=0.13 && \ + export PATCHELF_HASH=60c6aeadb673de9cc1838b630c81f61e31c501de324ef7f1e8094a2431197d09 && \ + export PATCHELF_DOWNLOAD_URL=https://github.com/NixOS/patchelf/archive && \ + manylinux-entrypoint /build_scripts/install-patchelf.sh + +COPY build_scripts/install-libxcrypt.sh /build_scripts/ +RUN export LIBXCRYPT_VERSION=4.4.26 && \ + export LIBXCRYPT_HASH=e8a544dd19171c1e6191a6044c96cc31496d781ba08b5a00f53310d001d58114 && \ + export LIBXCRYPT_DOWNLOAD_URL=https://github.com/besser82/libxcrypt/archive && \ + export PERL_ROOT=perl-5.34.0 && \ + export PERL_HASH=551efc818b968b05216024fb0b727ef2ad4c100f8cb6b43fab615fa78ae5be9a && \ + export PERL_DOWNLOAD_URL=https://www.cpan.org/src/5.0 && \ + manylinux-entrypoint /build_scripts/install-libxcrypt.sh + +COPY build_scripts/install-protobuf.sh /build_scripts/ +RUN export PROTOBUF_VERSION=3.17.3 && \ + export PROTOBUF_ROOT=protobuf-all-${PROTOBUF_VERSION} && \ + export PROTOBUF_HASH=77ad26d3f65222fd96ccc18b055632b0bfedf295cb748b712a98ba1ac0b704b2 && \ + export PROTOBUF_DOWNLOAD_URL=https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOBUF_VERSION} && \ + manylinux-entrypoint /build_scripts/install-protobuf.sh + +FROM runtime_base AS build_base +COPY build_scripts/install-build-packages.sh /build_scripts/ +RUN manylinux-entrypoint /build_scripts/install-build-packages.sh + + +FROM build_base AS build_git +COPY build_scripts/build-git.sh /build_scripts/ +RUN export GIT_ROOT=git-2.33.1 && \ + export GIT_HASH=02047f8dc8934d57ff5e02aadd8a2fe8e0bcf94a7158da375e48086cc46fce1d && \ + export GIT_DOWNLOAD_URL=https://www.kernel.org/pub/software/scm/git && \ + manylinux-entrypoint /build_scripts/build-git.sh + + +FROM build_base AS build_swig +COPY build_scripts/build-swig.sh /build_scripts/ +RUN export SWIG_ROOT=swig-4.0.2 && \ + export SWIG_HASH=d53be9730d8d58a16bf0cbd1f8ac0c0c3e1090573168bfa151b01eb47fa906fc && \ + export SWIG_DOWNLOAD_URL=https://sourceforge.net/projects/swig/files/swig/${SWIG_ROOT} && \ + export PCRE_ROOT=pcre-8.45 && \ + export PCRE_HASH=4e6ce03e0336e8b4a3d6c2b70b1c5e18590a5673a98186da90d4f33c23defc09 && \ + export PCRE_DOWNLOAD_URL=https://sourceforge.net/projects/pcre/files/pcre/8.45 && \ + manylinux-entrypoint /build_scripts/build-swig.sh + + +FROM build_base AS build_cpython +COPY build_scripts/build-sqlite3.sh /build_scripts/ +RUN export SQLITE_AUTOCONF_ROOT=sqlite-autoconf-3360000 && \ + export SQLITE_AUTOCONF_HASH=bd90c3eb96bee996206b83be7065c9ce19aef38c3f4fb53073ada0d0b69bbce3 && \ + export SQLITE_AUTOCONF_DOWNLOAD_URL=https://www.sqlite.org/2021 && \ + manylinux-entrypoint /build_scripts/build-sqlite3.sh + +COPY build_scripts/build-openssl.sh /build_scripts/ +RUN export OPENSSL_ROOT=openssl-1.1.1l && \ + export OPENSSL_HASH=0b7a3e5e59c34827fe0c3a74b7ec8baef302b98fa80088d7f9153aa16fa76bd1 && \ + export OPENSSL_DOWNLOAD_URL=https://www.openssl.org/source && \ + manylinux-entrypoint /build_scripts/build-openssl.sh + +COPY build_scripts/build-cpython.sh /build_scripts/ + + +FROM build_cpython AS build_cpython36 +COPY build_scripts/cpython-pubkeys.txt /build_scripts/cpython-pubkeys.txt +RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.6.15 + + +FROM build_cpython AS build_cpython37 +COPY build_scripts/cpython-pubkeys.txt /build_scripts/cpython-pubkeys.txt +RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.7.12 + + +FROM build_cpython AS build_cpython38 +COPY build_scripts/ambv-pubkey.txt /build_scripts/cpython-pubkeys.txt +RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.8.12 + + +FROM build_cpython AS build_cpython39 +COPY build_scripts/ambv-pubkey.txt /build_scripts/cpython-pubkeys.txt +RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.9.8 + + +FROM build_cpython AS all_python +COPY build_scripts/install-pypy.sh /build_scripts/install-pypy.sh +COPY build_scripts/pypy.sha256 /build_scripts/pypy.sha256 +RUN manylinux-entrypoint /build_scripts/install-pypy.sh 3.7 7.3.7 +RUN manylinux-entrypoint /build_scripts/install-pypy.sh 3.8 7.3.7 +COPY --from=build_cpython36 /opt/_internal /opt/_internal/ +COPY --from=build_cpython37 /opt/_internal /opt/_internal/ +COPY --from=build_cpython38 /opt/_internal /opt/_internal/ +COPY --from=build_cpython39 /opt/_internal /opt/_internal/ +RUN hardlink -cv /opt/_internal + + +FROM runtime_base +COPY --from=build_git /manylinux-rootfs / +COPY --from=build_swig /manylinux-rootfs / +COPY --from=build_cpython /manylinux-rootfs / +COPY --from=all_python /opt/_internal /opt/_internal/ +COPY build_scripts/finalize.sh \ + build_scripts/update-system-packages.sh \ + build_scripts/python-tag-abi-tag.py \ + build_scripts/requirements3.6.txt \ + build_scripts/requirements3.7.txt \ + build_scripts/requirements3.8.txt \ + build_scripts/requirements3.9.txt \ + build_scripts/requirements-base-tools.txt \ + /build_scripts/ +COPY build_scripts/requirements-tools/* /build_scripts/requirements-tools/ +RUN manylinux-entrypoint /build_scripts/finalize.sh && rm -rf /build_scripts + +ENV SSL_CERT_FILE=/opt/_internal/certs.pem + +CMD ["/bin/bash"] + +#Build manylinux2014 docker image end + +ARG PYTHON_VERSION=3.7 +ARG TORCH_VERSION=1.10.0 +ARG OPSET_VERSION=14 +ARG INSTALL_DEPS_EXTRA_ARGS + +#Add our own dependencies +ADD scripts /tmp/scripts +RUN cd /tmp/scripts && \ + /tmp/scripts/manylinux/install_centos.sh && \ + /tmp/scripts/install_os_deps.sh -d gpu $INSTALL_DEPS_EXTRA_ARGS && \ + /tmp/scripts/install_python_deps.sh -d gpu -p 3.6 -h ${TORCH_VERSION} $INSTALL_DEPS_EXTRA_ARGS && \ + /tmp/scripts/install_python_deps.sh -d gpu -p 3.7 -h ${TORCH_VERSION} $INSTALL_DEPS_EXTRA_ARGS && \ + /tmp/scripts/install_python_deps.sh -d gpu -p 3.8 -h ${TORCH_VERSION} $INSTALL_DEPS_EXTRA_ARGS && \ + /tmp/scripts/install_python_deps.sh -d gpu -p 3.9 -h ${TORCH_VERSION} $INSTALL_DEPS_EXTRA_ARGS && \ + rm -rf /tmp/scripts + +# remove protobuf to prevent ambiguity which is used for onnxruntime build +RUN rm -fr /usr/local/bin/protoc \ + /usr/local/libproto* \ + /usr/local/include/google \ + /usr/local/lib/pkgconfig/protobuf* + +ARG BUILD_UID=1001 +ARG BUILD_USER=onnxruntimedev +RUN adduser --uid $BUILD_UID $BUILD_USER +WORKDIR /home/$BUILD_USER +USER $BUILD_USER +ENV PATH /usr/local/gradle/bin:/usr/local/dotnet:$PATH +ENV ORTMODULE_ONNX_OPSET_VERSION=$OPSET_VERSION