Merge remote-tracking branch 'upstream/main' into rhoai-3.2

Author: CFSNM

tests/common/support/environment.go

@@ -58,6 +58,7 @@ const (
 	storageBucketName            = "AWS_STORAGE_BUCKET"
 	storageBucketMnistDir        = "AWS_STORAGE_BUCKET_MNIST_DIR"
 	storageBucketFashionMnistDir = "AWS_STORAGE_BUCKET_FASHION_MNIST_DIR"
+	storageBucketOsftDir         = "AWS_STORAGE_BUCKET_OSFT_DIR"
 
 	// Name of existing namespace to be used for test
 	testNamespaceNameEnvVar = "TEST_NAMESPACE_NAME"
@@ -185,6 +186,11 @@ func GetStorageBucketFashionMnistDir() (string, bool) {
 	return storage_bucket_fashion_mnist_dir, exists
 }
 
+func GetStorageBucketOsftDir() (string, bool) {
+	storage_bucket_osft_dir, exists := os.LookupEnv(storageBucketOsftDir)
+	return storage_bucket_osft_dir, exists
+}
+
 func GetPipIndexURL() string {
 	return lookupEnvOrDefault(pipIndexURL, "https://pypi.python.org/simple")
 }

tests/trainer/kubeflow_sdk_test.go

@@ -20,10 +20,17 @@ import (
 	"testing"
 
 	. "github.com/opendatahub-io/distributed-workloads/tests/common"
+	support "github.com/opendatahub-io/distributed-workloads/tests/common/support"
 	sdktests "github.com/opendatahub-io/distributed-workloads/tests/trainer/sdk_tests"
 )
 
 func TestKubeflowSdkSanity(t *testing.T) {
 	Tags(t, Sanity)
 	sdktests.RunFashionMnistCpuDistributedTraining(t)
 }
+
+// TestOsftTrainingHubMultiNodeMultiGPU tests OSFT training using TrainingHubTrainer
+func TestOsftTrainingHubMultiNodeMultiGPU(t *testing.T) {
+	Tags(t, KftoCuda, MultiNodeMultiGpu(2, support.NVIDIA, 1)) // TODO: may need to be updated once https://issues.redhat.com/browse/RHOAIENG-30719 and https://issues.redhat.com/browse/RHOAIENG-24552 are resolved
+	sdktests.RunOsftTrainingHubMultiGpuDistributedTraining(t)
+}