TensorFlow-Keras

This section describes how to install and run TensorFlow and Keras, and how to install TensorFlow, Keras and Horovod to perform distributed learning.

Note

Starting with TensorFlow 2, Keras is included in TensorFlow. This means that installing TensorFlow 2 will also enable you to use Keras. The TensorFlow and Keras installation steps described here are the same as the TensorFlow steps, except that the sample you run uses Keras.

Running TensorFlow-Keras on a single node

Precondition

• Replace grpname with your own ABCI group.
• The Python virtual environment should be created in the home or group area so that it can be referenced by interactive nodes and each compute node.
• The sample program should be saved in the home or group area so that it can be referenced by interactive nodes and each compute node.

Installation

Here are the steps to create a Python virtual environment and install TensorFlow and Keras into the Python virtual environment.

[username@es1 ~]$ qrsh -g grpname -l rt_G.small=1 -l h_rt=1:00:00
[username@g0001 ~]$ module load gcc/9.3.0 python/3.8/3.8.7 cuda/11.0/11.0.3 cudnn/8.0/8.0.5
[username@g0001 ~]$ python3 -m venv ~/venv/tensorflow-keras
[username@g0001 ~]$ source ~/venv/tensorflow-keras/bin/activate
(tensorflow-keras) [username@g0001 ~]$ pip3 install --upgrade pip setuptools
(tensorflow-keras) [username@g0001 ~]$ pip3 install tensorflow==2.4.1

With the installation, you can use TensorFlow and Keras next time you want to use it by simply loading the module and activating the Python virtual environment, as follows.

[username@g0001 ~]$ module load gcc/9.3.0 python/3.8/3.8.7 cuda/11.0/11.0.3 cudnn/8.0/8.0.5
[username@g0001 ~]$ source ~/venv/tensorflow-keras/bin/activate

Execution

The following shows how to execute the TensorFlow sample program mnist_convnet.py in the case of an interactive job and a batch job.

Run as an interactive job

[username@es1 ~]$ qrsh -g grpname -l rt_G.small=1 -l h_rt=1:00:00
[username@g0001 ~]$ module load gcc/9.3.0 python/3.8/3.8.7 cuda/11.0/11.0.3 cudnn/8.0/8.0.5
[username@g0001 ~]$ source ~/venv/tensorflow-keras/bin/activate
(tensorflow-keras) [username@g0001 ~]$ git clone https://github.com/keras-team/keras-io.git
(tensorflow-keras) [username@g0001 ~]$ python3 keras-io/examples/vision/mnist_convnet.py

Run as a batch job

Save the following job script as a run.sh file.

#!/bin/sh

#$ -l rt_G.small=1
#$ -j y
#$ -cwd

source /etc/profile.d/modules.sh
module load gcc/9.3.0 python/3.8/3.8.7 cuda/11.0/11.0.3 cudnn/8.0/8.0.5
source ~/venv/tensorflow-keras/bin/activate
git clone https://github.com/keras-team/keras-io.git
python3 keras-io/examples/vision/mnist_convnet.py
deactivate

Submit a saved job script run.sh as a batch job with the qsub command.

[username@es1 ~]$ qsub -g grpname run.sh
Your job 1234567 ('run.sh') has been submitted

Running TensorFlow on multiple nodes

Precondition

• Replace grpname with your own ABCI group.
• The Python virtual environment should be created in the home or group area so that it can be referenced by interactive nodes and each compute node.
• The sample program should be saved in the home or group area so that it can be referenced by interactive nodes and each compute node.

Installation

Here are the steps to create a Python virtual environment and install TensorFlow, Keras and Horovod into the Python virtual environment.

[username@es1 ~]$ qrsh -g grpname -l rt_G.small=1 -l h_rt=1:00:00
[username@g0001 ~]$ module load gcc/9.3.0 python/3.8/3.8.7 openmpi/4.0.5 cuda/11.0/11.0.3 cudnn/8.0/8.0.5 nccl/2.8/2.8.4-1
[username@g0001 ~]$ python3 -m venv ~/venv/tensorflow-keras+horovod
[username@g0001 ~]$ source ~/venv/tensorflow-keras+horovod/bin/activate
(tensorflow-keras+horovod) [username@g0001 ~]$ pip3 install --upgrade pip setuptools
(tensorflow-keras+horovod) [username@g0001 ~]$ pip3 install tensorflow==2.4.1
(tensorflow-keras+horovod) [username@g0001 ~]$ HOROVOD_WITH_TENSORFLOW=1 HOROVOD_GPU_OPERATIONS=NCCL HOROVOD_NCCL_HOME=$NCCL_HOME pip3 install --no-cache-dir horovod==0.22.0

With the installation, you can use TensorFlow, Keras and Horovod next time you want to use it by simply loading the module and activating the Python virtual environment, as follows.

[username@g0001 ~]$ module load gcc/9.3.0 python/3.8/3.8.7 openmpi/4.0.5 cuda/11.0/11.0.3 cudnn/8.0/8.0.5 nccl/2.8/2.8.4-1
[username@g0001 ~]$ source ~/venv/tensorflow-keras+horovod/bin/activate

Execution

The following shows how to execute a sample program tensorflow2_keras_mnist.py of TensorFlow with Horovod for distributed learning.

Run as an interactive job

In this example, using 4 GPUs in a compute node for distributed learning.

[username@es1 ~]$ qrsh -g grpname -l rt_G.large=1 -l h_rt=1:00:00
[username@g0001 ~]$ module load gcc/9.3.0 python/3.8/3.8.7 openmpi/4.0.5 cuda/11.0/11.0.3 cudnn/8.0/8.0.5 nccl/2.8/2.8.4-1
[username@g0001 ~]$ source ~/venv/tensorflow-keras+horovod/bin/activate
(tensorflow-keras+horovod) [username@g0001 ~]$ git clone -b v0.22.0 https://github.com/horovod/horovod.git
(tensorflow-keras+horovod) [username@g0001 ~]$ mpirun -np 4 -map-by ppr:4:node -mca pml ob1 python3 horovod/examples/tensorflow2/tensorflow2_keras_mnist.py

Run as a batch job

In this example, a total of 8 GPUs are used for distributed learning. 2 compute nodes are used, with 4 GPUs per compute node.

Save the following job script as a run.sh file.

#!/bin/sh

#$ -l rt_F=2
#$ -j y
#$ -cwd

source /etc/profile.d/modules.sh
module load gcc/9.3.0 python/3.8/3.8.7 openmpi/4.0.5 cuda/11.0/11.0.3 cudnn/8.0/8.0.5 nccl/2.8/2.8.4-1
source ~/venv/tensorflow-keras+horovod/bin/activate

git clone -b v0.22.0 https://github.com/horovod/horovod.git

NUM_GPUS_PER_NODE=4
NUM_PROCS=$(expr ${NHOSTS} \* ${NUM_GPUS_PER_NODE})

MPIOPTS="-np ${NUM_PROCS} -map-by ppr:${NUM_GPUS_PER_NODE}:node -mca pml ob1 -mca btl self,tcp -mca btl_tcp_if_include bond0"

mpirun ${MPIOPTS} python3 horovod/examples/tensorflow2/tensorflow2_keras_mnist.py

deactivate

Submit a saved job script run.sh as a batch job with the qsub command.

[username@es1 ~]$ qsub -g grpname run.sh
Your job 1234567 ('run.sh') has been submitted