QuickStart TensorFlow

Tensorflow is a symbolic math library based on dataflow and differentiable programming.

---

Use Global Protect VPN: whether on or off-campus

In top menu bar access (globe icon)
Be sure vpn-groups selected when you connect.

Start an SSH session:

ssh your_netid@hpc.kennesaw.edu

To test or develop, reserve a node (for non-gpu)

 $  qsub -I -q batch -l nodes=1:ppn=24,walltime=1:00:00

To test or develop, reserve a node (gpu)

 $  qsub -I -q gpuq -l nodes=1:ppn=24:gpus=2,walltime=1:00:00

---

Due to the variety of ways TensorFlow is used, it is now advised to build your own Python virtual environment to access a recent copy of TensorFlow

You will want to utilize the Anaconda module to create a Conda environment.

From your Conda environment, you can have access to TensorFlow with GPU support that should work without a GPU.

Once created, to use the environment in an interactive session or from within a PBS job submission script, you will need to load Anaconda and activate your new TensorFlow-aware conda environment.

Step by step instructions have been prepared by the KSU Sysadmins on the KSU HPCdocs wiki: TensorFlow Example

---

Write a job submission script (run_tf2.pbs)

 Check the PBS file example at hpcdocs.

Submit your job to the scheduler

Use the qsub sumission example at the bottome of the PBS file example.

---

Tensorflow tutorials and guides - https://www.tensorflow.org/tutorials