From 5daa029fdbd46bbbafb17612a7353099f6bbd988 Mon Sep 17 00:00:00 2001 From: Travis Addair Date: Mon, 9 Nov 2020 13:40:11 -0800 Subject: [PATCH] Use mxnet==1.5.0 for CPU Dockerfile until issues with 1.6.0 and 1.7.0 are resolved (#2426) Signed-off-by: Travis Addair --- Dockerfile.cpu | 4 ++-- build-docker-images.sh | 8 ++++---- 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/Dockerfile.cpu b/Dockerfile.cpu index 13721fb144..43f67c0b81 100644 --- a/Dockerfile.cpu +++ b/Dockerfile.cpu @@ -3,7 +3,7 @@ FROM ubuntu:18.04 ENV TENSORFLOW_VERSION=2.3.0 ENV PYTORCH_VERSION=1.6.0 ENV TORCHVISION_VERSION=0.7.0 -ENV MXNET_VERSION=1.6.0 +ENV MXNET_VERSION=1.5.0 # Python 3.7 is supported by Ubuntu Bionic out of the box ARG python=3.7 @@ -57,7 +57,7 @@ RUN mkdir /tmp/openmpi && \ rm -rf /tmp/openmpi # Install Horovod -RUN HOROVOD_WITH_TENSORFLOW=1 HOROVOD_WITH_PYTORCH=1 HOROVOD_WITH_MXNET=1 \ +RUN MAKEFLAGS="-j1" HOROVOD_WITH_TENSORFLOW=1 HOROVOD_WITH_PYTORCH=1 HOROVOD_WITH_MXNET=1 \ pip install --no-cache-dir horovod # Install OpenSSH for MPI to communicate between containers diff --git a/build-docker-images.sh b/build-docker-images.sh index f573fa6355..68bb9ec919 100755 --- a/build-docker-images.sh +++ b/build-docker-images.sh @@ -19,7 +19,7 @@ function build_one() fi tag=horovod-build-py${py}-${device}:$(date +%Y%m%d-%H%M%S) - docker build -f Dockerfile.${device} -t ${tag} --build-arg python=${py} --no-cache . + docker build -f Dockerfile.${device} -t ${tag} --build-arg python=${py} . horovod_version=$(docker run --rm ${tag} pip show horovod | grep Version | awk '{print $2}') tensorflow_version=$(docker run --rm ${tag} pip show ${tensorflow_pkg} | grep Version | awk '{print $2}') pytorch_version=$(docker run --rm ${tag} pip show torch | grep Version | sed 's/+/ /g' | awk '{print $2}') @@ -30,12 +30,12 @@ function build_one() } # clear upstream images, ok to fail if images do not exist -docker rmi $(cat Dockerfile.cpu | grep FROM | awk '{print $2}') || true -docker rmi $(cat Dockerfile.gpu | grep FROM | awk '{print $2}') || true +# docker rmi $(cat Dockerfile.cpu | grep FROM | awk '{print $2}') || true +# docker rmi $(cat Dockerfile.gpu | grep FROM | awk '{print $2}') || true # build for cpu and gpu build_one 3.7 cpu -build_one 3.7 gpu +# build_one 3.7 gpu # print recent images docker images horovod/horovod