|
| 1 | +# Steps to generate a container with Intel® Optimization for TensorFlow |
| 2 | + |
| 3 | +This guide will help you generate a container with Intel's cpx launch release candidate. |
| 4 | + |
| 5 | +## Steps: |
| 6 | + |
| 7 | +1. Clone intel-tensorflow cpx launch branch: |
| 8 | + |
| 9 | + ``` |
| 10 | + $ git clone https://github.com/Intel-tensorflow/tensorflow.git --branch=bf16/base --single-branch |
| 11 | + $ cd tensorflow |
| 12 | + $ git checkout 34305e8bf7ad4e7ffb9339b434385e3b28896924 |
| 13 | + # Run "git log" and check for the right git hash |
| 14 | + ``` |
| 15 | + |
| 16 | +2. Go to the directory that has Intel mkl docker files: |
| 17 | + |
| 18 | + ``` |
| 19 | + $ cd tensorflow/tools/ci_build/linux/mkl/ |
| 20 | + ``` |
| 21 | + |
| 22 | +3. Run build-dev-container.sh by passing the following env parameters: |
| 23 | + |
| 24 | + ``` |
| 25 | + $ env ROOT_CONTAINER=tensorflow/tensorflow \ |
| 26 | + ROOT_CONTAINER_TAG=devel \ |
| 27 | + TF_DOCKER_BUILD_DEVEL_BRANCH=34305e8bf7ad4e7ffb9339b434385e3b28896924 \ |
| 28 | + TF_REPO=https://github.com/Intel-tensorflow/tensorflow \ |
| 29 | + TF_DOCKER_BUILD_VERSION=tensorflow-2.2-bf16-nightly \ |
| 30 | + BUILD_SKX_CONTAINERS=yes \ |
| 31 | + BUILD_TF_V2_CONTAINERS=yes \ |
| 32 | + BUILD_TF_BFLOAT16_CONTAINERS=yes \ |
| 33 | + BAZEL_VERSION= \ |
| 34 | + ENABLE_DNNL1=yes \ |
| 35 | + ENABLE_SECURE_BUILD=yes \ |
| 36 | + ./build-dev-container.sh > ./container_build.log |
| 37 | + ``` |
| 38 | + |
| 39 | +4. Open a second terminal session at the same location and run `tail -f container_build.log` to monitor container build progress |
| 40 | + or wait until the build finishes and then open the log file <container_build.log> ... |
| 41 | + |
| 42 | + ``` |
| 43 | + INFO: Build completed successfully, 18811 total actions. |
| 44 | + ``` |
| 45 | + |
| 46 | + Below output indicates that the container has intel-optimized tensorflow: |
| 47 | + |
| 48 | + ``` |
| 49 | + PASS: MKL enabled test in <intermediate container name> |
| 50 | + ``` |
| 51 | + |
| 52 | +5. Check if the image was built successfully and tag it: |
| 53 | + |
| 54 | + ``` |
| 55 | + $ docker images |
| 56 | + intel-mkl/tensorflow:tensorflow-2.2-bf16-nightly-avx512-devel-mkl |
| 57 | + |
| 58 | + $ docker tag intel-mkl/tensorflow:tensorflow-2.2-bf16-nightly-avx512-devel-mkl intel/intel-optimized-tensorflow:tensorflow-2.2-bf16-nightly |
| 59 | + ``` |
| 60 | + |
| 61 | +6. Run the image in privileged mode and install OpenMPI, OpenSSH and Horovod: |
| 62 | + Example of docker run command: |
| 63 | + |
| 64 | + ``` |
| 65 | + $ docker run --init --privileged -it --env <Proxy setup or anything else> -v <mount dir> --name container_name <imageid> /bin/bash |
| 66 | + ``` |
| 67 | + |
| 68 | + Install Open MPI |
| 69 | + |
| 70 | + ``` |
| 71 | + $ apt-get clean && apt-get update -y |
| 72 | + $ apt-get install -y --no-install-recommends --fix-missing openmpi-bin openmpi-common libopenmpi-dev |
| 73 | +
|
| 74 | + # Check OpenMPI installation: |
| 75 | + $ mpirun --version |
| 76 | + # You should see the following message: |
| 77 | + mpirun (Open MPI) 2.1.1 |
| 78 | + ``` |
| 79 | + Install OpenSSH for MPI to communicate between containers |
| 80 | + ``` |
| 81 | + |
| 82 | + $ apt-get install -y --no-install-recommends --fix-missing openssh-client openssh-server libnuma-dev |
| 83 | + $ mkdir -p /var/run/sshd |
| 84 | + # Allow OpenSSH to talk to containers without asking for confirmation |
| 85 | + $ cat /etc/ssh/ssh_config | grep -v StrictHostKeyChecking > /etc/ssh/ssh_config.new |
| 86 | + $ echo " StrictHostKeyChecking no" >> /etc/ssh/ssh_config.new |
| 87 | + $ mv /etc/ssh/ssh_config.new /etc/ssh/ssh_config |
| 88 | + ``` |
| 89 | + Install Horovod |
| 90 | + |
| 91 | + ``` |
| 92 | + $ HOROVOD_WITH_TENSORFLOW=1 |
| 93 | + $ python3 -m pip install --no-cache-dir horovod==0.19.1 |
| 94 | + ``` |
| 95 | + |
| 96 | + If Horovod installation was successful you will see the following message: |
| 97 | + |
| 98 | + ``` |
| 99 | + Successfully installed cffi-1.14.0 cloudpickle-1.4.1 horovod-0.19.1 psutil-5.7.0 pycparser-2.20 pyyaml-5.3.1 |
| 100 | + ``` |
| 101 | + |
| 102 | + Check Horovod installation: |
| 103 | + |
| 104 | + ``` |
| 105 | + $ python -c "import tensorflow as tf; import horovod.tensorflow as hvd;" |
| 106 | + ``` |
| 107 | + You should not see an error. |
| 108 | +
|
| 109 | +7. Save this image: |
| 110 | + |
| 111 | + ``` |
| 112 | + $ exit |
| 113 | + $ docker ps -a |
| 114 | + $ docker commit [container ID] intel/intel-optimized-tensorflow:tensorflow-2.2-bf16-nightly |
| 115 | + ``` |
| 116 | + |
| 117 | +Substitute this docker image when using the parameter `--docker-image` in running benchmarks [`launch_benchmark.py`](/benchmarks/launch_benchmark.py). |
| 118 | + |
| 119 | + |
0 commit comments