* SDK - Client - Added a way to set experiment name using environment variables This is useful for launching notebooks or pipeline files that submit themselves for execution. * Switched to subprocess.run which supports env * Setting the environment variable differently Looks like `subprocess.run` uses `PATH` to search for the program. * Convert return code to string * Changed the way the experiment name is being set * Changed how the notebook installs the SDK Notebook is overriding the SDK that's being tested. * Not installing the KFP SDK package * Removed the experiment_name from samples and configs. * Changed the SDK installation lines in samples Otherwise the sample tests do not correctly test the new SDK code. |
||
|---|---|---|
| .. | ||
| mpi | ||
| standalonejob | ||
| README.md | ||
| arena.yaml | ||
| arena_launcher_rbac.yaml | ||
README.md
Arena demo
There are a series of examples about how to build deeplearning models with Arena. These demos will show how to run a pipeline standalone Job, MPI Job, TFJob(PS mode) and TensorFlow Estimator Job.
Setup
- Install the arena
kubectl create -f https://raw.githubusercontent.com/kubeflow/pipelines/eb830cd73ca148e5a1a6485a9374c2dc068314bc/samples/arena-samples/arena.yaml
- Add addtional RBAC role to service account
pipeline-runner
kubectl create -f https://raw.githubusercontent.com/kubeflow/pipelines/eb830cd73ca148e5a1a6485a9374c2dc068314bc/samples/arena-samples/arena_launcher_rbac.yaml
Demos
- Standalone Job
- MPI Job
- TensorFlow Estimator Job
- TFJob