Launch TensorFlow Docker Application

This guide shows how to run the TensorFlow Docker application. Sign in to AAC if you have not already.

Select application

Click Applications, then select TensorFlow. In the Select An Application pop-up, choose the TensorFlow version with container type docker.

Note

In this example we use TensorFlow 2-10 ROCm 5-4-1 with docker.

New workload

Click New Workload at the top right.

Select team

If you have more than one team, select one in the pop-up and click Launch. Click Next to continue.

Select resources

Set the number of GPUs (max 8) and maximum allowed runtime. Click Next. Select the cluster and queue for the job (e.g. 1CN128C8G2H_2IB_MI210_SLES15 (Pre-emptible) - AAC Plano), then click Next.

Review workload submission

Review the configuration and click Run Workload.

When the queue is available, the status changes to Running. Click the running workload to open it.

Use the SYSLOG, STDOUT, and STDERR tabs to view logs and output. A token appears in STDOUT (in yellow). Copy the token.

Interactive endpoints

When interactive endpoints are enabled, click Connect to open ML Studio (JupyterLab).

Paste the token in the Password or token field and click Login.

You can use JupyterLab for Python development.

Click Terminal to open a terminal. To run the benchmark, enter:

python3 /root/benchmarks/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py --model=resnet50 --num_gpus=8 --batch_size=256 --num_batches=100 --print_training_accuracy=True --variable_update=parameter_server --local_parameter_device=gpu

Collect Performance metrics as needed.

When you are done, close JupyterLab.

Finish workload

Click Finish Workload.

Download logs

After the workload finishes, download logs from the STDOUT tab via Download Logs.