in dags/mlcompass/configs/simple_config.py [0:0]
def get_simple_config():
set_up_cmds = (
"set +x",
"echo {{params.commit_sha}}",
"pip install -U pip",
"pip install jax[tpu] -f https://storage.googleapis.com/jax-releases/libtpu_releases.html",
)
run_model_cmds = (
"set +x",
"echo {{params.commit_sha}}",
"ls -ltrh /dev/accel*",
"python3 -c 'import jax; print(jax.device_count()); print(jax.numpy.add(1,1))'",
)
job_test_config = test_config.TpuVmTest(
test_config.Tpu(
version=TpuVersion.V4,
cores=8,
runtime_version=RuntimeVersion.TPU_UBUNTU2204_BASE.value,
reserved=False,
network="default",
subnetwork="default",
),
test_name="simple-jax-code",
set_up_cmds=set_up_cmds,
run_model_cmds=run_model_cmds,
timeout=datetime.timedelta(minutes=60),
task_owner=test_owner.ORTI_B,
num_slices=1,
)
project_name = Project.CLOUD_ML_AUTO_SOLUTIONS.value
job_gcp_config = gcp_config.GCPConfig(
project_name=project_name,
zone=Zone.US_CENTRAL2_B.value,
dataset_name=metric_config.DatasetOption.XLML_DATASET,
dataset_project=project_name,
composer_project=project_name,
)
return task.run_queued_resource_test(
task_test_config=job_test_config,
task_gcp_config=job_gcp_config,
)