CANDLE tests

These should be runnable interactively or via Jenkins.

The Jenkins project is here: https://jenkins-ci.cels.anl.gov/job/CANDLE-CR

First, make sure your desired Python installation is in PATH and PYTHONPATH as needed.

Second, simply run the test you want, e.g.:

or to run all tests:

Each test has a separate run directory (RUNDIR), but this is overwritten from run to run. We assume the numbered/dated output logs in Jenkins will capture everything needed to debug a historical test run.

Each test is associated with a TOKEN based on its directory name ("NT3", "P1B1", etc.). A corresponding setup-TOKEN.sh may be created that will automatically be sourced by the scripts.

check-count.sh

Check counts of things like files

check-epoch.sh

Output checker for epochs. Knows how to find epoch output for Keras and PyTorch

check-output.sh

Output checker

check_tf.py

Keras/TensorFlow settings reporter

checkout-benchmarks.sh

Checks out or updates Benchmarks into a subdirectory

find-benchmarks.sh

Find the directory Benchmarks

py-settings.sh

Report Python settings

setup-tests.sh

Sets up Python, PATH, etc. Needed only once for a series of tests.

setup-test.sh

Sets up a particular Benchmark

utils.sh

Generic shell scripting tools

cr-nt3-1.sh

Test checkpoint/restart for NT3

cr-p1b1-torch-1.sh

Test checkpoint/restart for P1B1/PyTorch

cr-1.sh

Internal test pattern script. Runs any Benchmark and applies a sequence of CANDLE checkpoint/restart arguments.

1. Duplicate cr-1.sh and cr-list-1.sh

2. Change the commands in the cr-1.sh to do what you want

3. Check the output with the check-*.sh scripts or add your own tests